Storage Concepts Storing and Managing Digital Data

Storage Concepts
Storing and Managing Digital Data
Hitachi Data Systems Academy
Edition 1.3
Storage Concepts
Storage Concepts
Published by
HDS Academy, Hitachi Data Systems
Corporate Headquarters
2845 Lafayette Street
Santa Clara, California 95050-2639 USA
www.HDS.com community.HDS.com
Americas: +1 866 374 5822 or [email protected]
Europe, Middle East and Africa: +44 (0) 1753 618000 or [email protected]
HITACHI is a trademark or registered trademark of Hitachi, Ltd.
form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise,
LIMIT OF LIABILITY/DISCLAIMER OF WARRANTY: THE PUBLISHER AND THE AUTHORS MAKE NO

REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE ACCURACY OR COMPLETENESS
OF THE CONTENTS OF THIS WORK AND SPECIFICALLY DISCLAIM ALL WARRANTIES, INCLUDING
WITHOUT LIMITATION WARRANTIES OF FITNESS FOR A PARTICULAR PURPOSE. NO WARRANTY
MAY BE CREATED OR EXTENDED BY SALES OR PROMOTIONAL MATERIALS. THE ADVICE AND
STRATEGIES CONTAINED HEREIN MAY NOT BE SUITABLE FOR EVERY SITUATION.
THIS WORK IS SOLD WITH THE UNDERSTANDING THAT THE PUBLISHER IS NOT ENGAGED IN
RENDERING LEGAL, ACCOUNTING OR OTHER PROFESSIONAL SERVICES. IF PROFESSIONAL
ASSISTANCE IS REQUIRED, THE SERVICES OF A COMPETENT PROFESSIONAL PERSON SHOULD BE
SOUGHT. NEITHER THE PUBLISHER NOR THE AUTHOR SHALL BE LIABLE FOR DAMAGES ARISING
HEREFROM. THE FACT THAT AN ORGANIZATION OR WEBSITE IS REFERRED TO IN THIS WORK
AS A CITATION AND/OR A POTENTIAL SOURCE OF FURTHER INFORMATION DOES NOT MEAN
THAT THE AUTHOR OR THE PUBLISHER ENDORSES THE INFORMATION THE ORGANIZATION
OR WEBSITE MAY PROVIDE OR RECOMMENDATIONS IT MAY MAKE. FURTHER, READERS
SHOULD BE AWARE THAT INTERNET WEBSITES LISTED IN THIS WORK MAY HAVE CHANGED OR
DISAPPEARED BETWEEN WHEN THIS WORK WAS WRITTEN AND WHEN IT IS READ.
Acknowledgements
Editor: Peter Manijak
Martin Stewart
Pavel Vild
Rebecca George
Copyeditor:
Graphical work:
Vojtech Marek

Tina Pankievich

Consultants and specialists:
Dr. Miroslav Kotrle

Radim Petrzela

Project managers:

Cecil Chamberlain
Jaroslav Fojtik
Produced and developed by Hitachi Data Systems Academy in cooperation

withMHM computer a.s.
Hitachi Data Systems Corporation 2012. All Rights Reserved.
ISBN-13: 978-0615656496 (HDS Academy, Hitachi Data Systems)
ISBN-10: 0615656498
Contents
Table of contents
Foreword11
Conventions used in the book
13
1. Brief introduction to data management and storage systems

Introduction to data management
General user data
Business sector data
State institutions data
Structured and unstructured data
New trends in data management cloud computing
Data and information
Levels of data processing
Hard drive partitioning
Volume
File system
Logical units on a storage system
Microcode
Data consistency
Data integrity
15
16
17
18
20
22
24
26
28
28
30
32
34
36
37
39
2. Storage components and technologies

41
Overview of disk array components
42
How does a hard drive work?
42
Parallel advanced technology attachment
47
Serial advanced technology attachment
47
Small computer systems interface
48
Serial attached SCSI
49
Solid-state drive (SSD) a hard drive with no platters
49
Connecting hard drives together (RAID levels and their function) 50
Data striping
51
Data mirroring
52
Data spanning
53
Parity
54
Just a Bunch of Discs (JBOD)

RAID-0
RAID-1
RAID-1+0
RAID-5 and RAID-6
RAID level range
Relative RAID performance
Software and hardware RAID
Spare disks
Storage system hard drive enclosures
Back end architecture
Fibre Channel Arbitrated Loop (FC-AL) architecture
SAS back end architecture in detail
Cache
Cache operations
Cache data protection
Front end architecture
LUN ownership
Controller load balancing
Fibre Channel ports and their configuration
iSCSI interface
Enterprise storage systems
54
55
56
56
57
58
59
60
62
66
69
71
73
75
78
79
80
82
86
87
88
89
3. Storage networking and security

Introduction to networks
Why do we need networks?
Network cables
Network members nodes
Storage networking topologies
Direct attached storage (DAS)
Storage area network (SAN)
Storage area network components
Storage area network over iSCSI interface
Network attached storage (NAS)
NAS implementation and converged infrastructure
Network protocols
Storage area network security
Network attached storage security
93
94
94
95
99
103
105
106
108
109
110
113
115
117
120
4. Business continuity and replication

Business continuity concepts
Business continuity planning
Business impact analysis
Risk assessment
Business continuity vs. disaster recovery
Disaster recovery plan
Disaster recovery support in storage systems
Backup and replication
Backup concepts
Backup based on tape libraries
Backup strategies
Backup optimization
Replication objectives
Overview of replication commands
In-system replication: logical unit cloning
Copy-on-write snapshots
Server clusters
Remote replication: logical unit mirroring
Synchronous remote replication
Asynchronous remote replication
Dynamic replication appliance
Remote replication and geoclusters
Three data center multi-target replication
Replicated data consistency
Continuous data protection (CDP)
Redundancy in basic infrastructure
Tests and drills of disaster recovery procedures
Disasters treatment overview
123
124
125
127
132
134
135
137
138
140
144
145
146
149
151
154
157
160
163
165
167
168
170
172
174
175
176
178
179
5. Virtualization of storage systems

What is virtualization?
Virtualization: area of application
Layers of virtualization
Virtualization benefits
VMware based server virtualization
Hyper-V based server virtualization
Virtualization with blade servers
183
184
186
190
191
192
194
195
VM hardware requirements and networking

Memory overcommitment in server virtualization
Thin provisioning
Thin provisioning limitations
Virtualization in storage system controller
Virtualization of a storage system: logical partitioning
Automated tiered storage
Network based virtualization
Cloud computing and storage
Virtualization: the cons
Future of virtualization
197
198
199
204
206
210
212
216
218
221
222
6. Archiving and file and content management

Introduction to archiving
Fixed content
Other differences between archiving and backup
Motivation for data archiving
The right format of archived data
Digital archive: object oriented storage system
Data objects and their components
Digital archive features
Digital archive accessibility
Digital archive compliance features
Metadata in digital archives
Scalability, hardware designing and migration
Integration of a digital archive
Digital archive cloud features
Policy based content management
Archiving and content management
Enterprise content management solutions
225
226
227
228
229
233
234
236
238
240
242
244
245
246
247
248
249
250
7. Storage system administration

Who is a storage system administrator?
Storage administrator tasks
Common operations with LUNs and cache
Monitoring of storage systems
Storage system management software
Integration of a new storage system
253
254
255
258
262
265
268
Effective allocation of resources and reporting

Security management
Tasks of a storage system implementer
269
269
270
8. Performance tuning, optimization and storage consolidation

Motivation for performance tuning
Introduction to performance tuning
What can be measured and what is the output?
Performance tuning in midrange storage systems
Enterprise storage performance basics
Software for performance tuning and automated optimization
Consolidation when and why?
Consolidation project
273
274
275
277
283
286
290
292
293
9. Business challenges
Business challenges
Data growth forecast
Advanced data classification and tiering
High availability challenges
Fast and cost-effective response to business growth
Compliance and security challenges
Power requirements and cooling
HDD and FAN power savings
Green data center
Total cost of ownership
297
298
299
303
307
308
309
311
313
314
316
Self-assessment test
319
Glossary
327
11
Foreword
Data. Its a word we all know and use regularly, but how does it impact our personal and professional lives? We live in an era of technology that places information at our fingertips. Technology empowers us
to create anything from digital photographs, films and symphonies to
advanced medical images and space systems. Data grows exponentially
and stretches further each day. It makes the highest achievements possible and connects us like never before.
But how do we store this ever growing data the countless files
created daily around the world? The answer depends on the data owner. Individuals need to store files in an easily accessible location and
maximize free space for new files. Small businesses face more complex
issues with the data they use and store, but it is critical to store their
data. Midsized businesses and large corporations face far greater concerns. Their information, or data, is a corporate asset that needs to be
stored, managed and protected plus there are government compliance regulations that impact the process.
For years, many IT professionals have seen data storage as a mystery.
Larger businesses employ dedicated storage administrators or contract
with professional services consultants to perform the more complex
procedures and manage the all important data. Most IT professionals
know a lot about servers but very little about storage systems. However,
these specialized systems are a modern day necessity for the storage
of data. In this regard, an IT administrator (or other responsible party)
must have at least elementary knowledge of common problems related
to data storage and storage systems.
This book covers the basic areas of data storage. Each chapter is written as an independent section, so you do not have to read the chapters in serial order to glean knowledge on a particular topic; feel free
to jump from one chapter to another and select only the chapters that
interest you.
12
The book is for anyone who would like to understand the cornerstones of data storage technology and the significance of basic terminology used in connection with storage systems. We recommended this
book for students or starter administrators, as well as experienced IT
workers who want to deepen their knowledge in data storage. Some
individuals do not work directly with storage systems but need to communicate with personnel in IT; the book is suitable for this audience as
well.
Our intent is not to provide a detailed manual for coping with the
technologies of individual manufacturers, concrete commands, scripts
and procedures, or to compare equipment from different vendors in
the market. We respectfully leave these next steps to you.
After reading this book, you should understand the basic rules of
data storage and get a systematic overview of modern tools and technologies. You will also be acquainted with the economic viewpoint of
technology operation, and for this reason, you will better understand
the essence of IT investments.
If the subject of data storage systems piques your interest and you
want to deepen your knowledge after reading this book, you can search
for additional information through data storage companies such as Hitachi Data Systems (www.hds.com).
We wish you a successful first step in the amazing world of data storage and storage systems.
13
Conventions used in the book

In this book, we have adopted various conventions to provide you
with clarity and an enjoyable reading experience as you explore the
concepts and architecture of storing and managing digital data. The
content is logically structured into nine chapters and written in simple language suitable for readers whose native language is not English.
Each chapter starts with an overview to introduce you to a new topic
and ends with a summary to help you consolidate the knowledge you
have gained. The text is supplemented with pictures and diagrams to
illustrate the content through examples. Throughout the book, you will
encounter practical hints marked with the Instructors Tip icon and
highlights marked with the Remember icon:
The Instructors Tip icon marks additional information that provides

you with a new perspective on a given topic. Exploring the topic from different points of view should make it easier for you to understand the complex principles and concepts of data management and storage systems.
The Remember icon accompanies summaries of what was discussed

in the preceding section. You should pay special attention to these summaries since they contain the most important information you should remember after reading this book
14
Brief introduction to data

management and storage systems
What are you going to learn in this chapter?
What kind of data we have
The importance of data for business
Data and information whats the difference?
How data looks from the perspective of a storage system
How to distinguish between physical and logical level
of data processing
The concepts of data consistency and data integrity
16
1 Brief introduction to data management and storage systems
Introduction to data management

The changing forms of data
Before we get to storage systems and their components, lets take

a closer look at data itself. There are not many people who realize that
whenever they take a picture with their camera or send a document by
email they bring large amounts of data into existence. This data needs
to be stored and backed up somewhere, which of course creates an
increasing demand for data storage. Few people nowadays are satisfied
with mere conservation of data on their personal computer hard drives.
More people want unlimited access to their data through the Internet
and the ability to share that data with their friends and families. This
common goal creates a lot of space for business, and thats why companies offer a wide range of services focused on data backup and sharing.
To get an idea of the amount of data handled nowadays, simply look at
social networks, media sharing websites and other similar sites.
However, common users arent the only ones generating data; institutions like banks, hospitals and telecommunication companies also
generate data. This is the point where it all gets interesting. Unlike common users, these companies are highly dependent on data, since data
is the crucial element that enables them to do their business. Imagine a
bank that loses Internet banking transaction data. Even if the bank lost
data collected in the past fifteen minutes, it would lose huge amounts
of money and face potentially devastating consequences.
In this chapter, well talk about data and its importance. Youll see
simple examples that should help you understand the depth and complexity of technology connected with data storage. To give this topic a
bit of structure, well divide data into subgroups and list characteristics
these subgroups possess. The division is not an exact science since the
boundaries between types of data are often blurred. However, this ap-
17
proach will help you become familiar with key terms that well look at
in detail later.
Common user data:
Photos uploaded to web galleries or social networks
Movies uploaded to services like youtube.com
Documents created by online applications like Google Docs
Email communication stored on the Internet
Data backed up to online storage such asMicrosoft Sky Drive
Personal web pages and blogs
Increasingly popular cloud systems and online applications
All this data can be considered important, at least from the users
point of view, and deserves careful and secure handling. However,
limited access to these types of data for example, during hardware maintenance usually doesnt create any financial loss to the
end user. If you cant access your photo gallery for an hour once a
year, it most likely wont cost you a lot of money. For large institutions that store ongoing financial transaction data, this disruption can
have sometimes devastating consequences. To prevent data loss, you
need to employ several basic concepts various techniques for data
backup and data replication. The key term in the storage technology is
redundancy. Data needs to be stored as several copies in several locations to ensure no disaster can affect its availability.
18
Figure 1.1 Services like YouTube and Facebook have to store huge amounts of data.
Basically, redudancy is achieved through techniques commonly called

data replication. Data can be replicated on the basis of in-system replication or remote replication to a storage system located in another town
or even country. The process of data replication is rather complex, and
there are many tools for creating duplicates or clones of data. Thats why
we dedicated a whole chapter to this topic. Aside from data replication,
theres also hard drive mirroring, which well discuss further when we talk
about RAID technology.
Business sector data:

Accounting, invoices and financial records
Databases that contain data about clients
Email communication
Digitalization of printed documents
Archiving
19
Data that contains information about clients personal information

or financial transactions plays a crucial role for business. Imagine a telecommunications company that cannot store data about phone calls,
sent text messages or data traffic. When this company loses such data,
its unable to invoice the customer and request payment for the provided services. A malfunction in the storage system could therefore mean
serious financial loss for the company. In this case, it is not enough to
have the data backed up several times. The concept of redundancy
doesnt solve this problem. On this level, its a matter of course that the
company doesnt lose any data that was stored before the storage system failure; the key is to ensure business continuity to bring all the applications back online as fast as possible. Each company that needs the
highest data availability possible also needs a disaster recovery plan.
This brings us to two other key storage systems terms: business continuity and disaster recovery.
An institution thats highly dependent on the availability of its data
must be prepared to deal with all possible risk situations, such as natural disasters, cyber attacks, failure of public infrastructure. For example,
imagine a fire destroys all infrastructures servers and storage systems
how long would it take to bring the essential components back online? How extensive would the loss of data be? All these criteria need to
be considered in a business continuity plan.
Figure 1.2 Business companies are dependent onthe properly working data storage system.
20
State institutions data:

Databases and other structured data
Audio and video records (CCTV, court records)
Hospital records
Confidential and classified data
Archiving and digitalization
With this type of data, we can demonstrate other aspects that need
to be taken into consideration when designing a storage solution for the
customer. We have already mentioned redundancy, business continuity
and disaster recovery. When it comes to legal requirements, however,
there are some other verticals that need to be considered. Private businesses and state institutions typically have apolicy that governs how
long data must be retained. For medical records, it can be several decades. The data from court trials also needs to be archived over a long
period of time. On the other hand, data from CCTV cameras and the
like do not have to be stored for longer than several weeks or months.
When we talk about data, we also talk about its lifecycle. When we have
a storage system, it must be able to retain data over a long time with the
possibility of easy data migration to other devices. Legally mandatory
data retention does not apply only to data generated by state institutions even companies are legally bound to keep records on their accounting for several years.
The term data lifecycle applies to all data that comes into existence,
while a data retention period applies only to certain types of data and is
governed by law.
Another thing to mention here is that long-term data retention is often inversely proportional to high data availability. This means we have
to store certain data for, lets say, forty years, but its fast availability
21
is not so important. Therefore, we do not have to store this data on

the fastest and most efficient hard drives we can use archiving storage systems that are based on cheaper and less sophisticated hardware
such as magnetic tape technology.
Figure 1.3 shows how data lifecycles work and what types of technology are used according to data availability demands. For most frequently accessed data, we will use the most powerful hardware, for
example the most up-to-date storage systems equipped with SSD disks
or fast SAS hard drives. For data that does not have to be available as
quickly, we can use less powerful and less expensive storage systems
with SATA hard drives. There are also special storage systems designed
especially for archiving; these systems use data compression and are
equipped with less powerful but more reliable hard drives. For legacy
data and backed up data, we can use offline tape systems. Data stored
on these devices cannot be accessed immediately, and it can take hours
to get data ready for use.
Data Lifecycle
High Performance
Storage Tier
Low Cost
Storage Tier
Online Archive
Storage Tier
Offline
Archive
Figure 1.3 Data lifecycle diagram.
22
We can illustrate how the data lifecycle works by using accounting as

an example. Every company is legally bound to keep records. Accounting data of the ongoing quarter is stored on their high peformance tier
because the accountants are working with this data frequently. After the
quarterly financial statement, the data is moved to a low cost storage tier
because it is not accessed or modified that often. Once there is a new
financial year, the old data is transferred to an online archive storage tier
and stays there for another five years, which is required by law. At this
stage, the data is still accessible online. After five years, the company migrates the legacy accounting data to the offline archive, most likely based
on tape technology. The data lifecycle is closely related to the changing
value of information over time.
Structured and unstructured data

What is the difference?
In addition to dividing the data according to purpose, its possible to

divide the data according to structure. Structured data can be stored
in rows and columns while unstructured data cannot. Structured data,
therefore, consists of mostly databases, such as client databases, balance sheets and phonebooks. Unstructured data refers to computerized information that either does not have a data model or does not
have one thats easily usable by a computer program. The term unstructured data distinguishes such information from data stored in a structured form in databases or annotated (tagged) in documents.
The ongoing information explosion has consumed huge amounts
of expensive and valuable resources, both human and technical. The
management of unstructured data is a major problem. To manage and
process structured data with clearly defined columns and rows (mostly
23
databases and tables) you can use software for analysis. Then people
can use this data for interpretation. As for managing and processing
unstructured data, you need more sophisticated software able to work
with just one file type (for example, with MP3 files or Microsoft Word
documents). The computer analysis of unstructured data does not always provide satisfying results and there is often a need for human factor intervention. In other words, there must be a person to organize,
manage, process and analyze this data. Some examples of unstructured
data are:
Medical images such as MRI scans and x-rays
Photographs and diagrams
Digital documents such as check images and contracts
Electronic assets such as presentations, emails and design documents
To have a general idea of the structured and unstructured data ratio,
refer to Figure 1.4, which illustrates the exponentially growing demand
for data storage.
(TB)
Unstructured Data
Database
Email
Figure 1.4 Structured and ustructured data ratio, growing demands on data
storage capacity, HDS statistics.
24
In this section, we discussed several types of data. You learned that different levels of importance can be attributed to data, at least from the perspective of business and storage systems. The trick is ensuring that data
is safely stored and available for creation of several copies. Whenever you
send an email or upload a picture to a web gallery, you create at least two,
maybe even three or four copies, because the service provider cannot afford losing your data just because of technical failure. This, of course, demonstrates that the need for storage space is increasing, because all data
must be duplicated multiple times. This is called redundancy and it is a
key word that will be mentioned many times throughout this book. Other
terms worth remembering are business continuity and disaster recovery.
We will talk about these in detail later.
Data plays a crucial role in business nowadays. It is said that if a bank

computer system failed for two or three days and the bank could not store
transactions, it would cause a panic and the bank would most likely collapse. This is just one example of how important data is!
New trends in data management: cloud computing

To better illustrate the development of storage systems and the demand for more and more storage capacity, we can take a look at cloud
computing, increasingly popular among companies and users. You probably know that on the Internet there are various services that offer disk
space, often for free, with the possibility to upgrade to a larger storage
area for little money. You can then use this storage space as you wish
to back up your photos, documents and applications, and access them
from any computer connected to the Internet. This type of service can
be seen as apredecessor to newly emerging cloud computing. Cloud
computing goes even further it not only offers disk space at your disposal, but also the possibility to use the computing power of the data
center to run your online applications. The idea is that you do not have
25
to install the software on your computer but you can use online software on the cloud. The advantages of cloud computing are clear:
The user does not have to worry about licensing and software
costs.
The user has the same working environment on any computer in
any location that is connected to the Internet.
The user does not need powerful hardware to run certain applications; all the data processing and computing is done by a
remote computer that is part of the cloud solution.
All online applications are automatically updated.
Systems are scalable and pay as you go you can buy the exact
amount of services you need.
There is a low risk of data loss.
Figure 1.5 An example of cloud technology, iCloud solutions provided by Apple. Among its
competitors is Google, which offers Google Apps such as Google Docs.
26
On the other hand, there are disadvantages the biggest problems

can be the dependency on the solution provider, possible security risks
and Internet connection issues. Furthermore, the applications available
on cloud solutions usually do not offer the full functionality of the software you install on your computer, especially because of limitations in
Internet connection speed and latency. This whole area is not very old
and only the future will show if this type of data and application access
prevails.
Data and information
Why should we distinguish the two?
In the last section, we discussed what forms data can take and its
importance to businesses. Before we continue, lets define the term
data itself and make a clear distinction between data and information. This should help you understand more clearly the way data is
handled in computers and storage systems. We can say that data is a
physical and written representation of information, knowledge and
results of any real world observation.
From an information technology point of view, data:
Is a succession of written characters, which can be represented by numbers, letters or symbols. These characters may seem
strange or meaningless at first glance, but they serve to organize
the data.
Does not have to bear any useful information.
Must be processed, organized and structured in order to extract
information.
Is easily processed by modern electronic devices. Its the language of computers and storage systems.
27
On the other hand, information is a more complex term.

Information:
Is a meaningful interpretation of certain parts of data
Does not have to be written in characters
Cannot be easily processed by electronics devices
Has value that is not linear, but constantly changing
Is a commodity and can be bought or sold
When we compare these two definitions, we see that information is
superior to data. We do not store information to get data, but we store
data to preserve and pass the information on. We can also see that the
value of information is different from person to person. The same set
of data can be interpreted differently by different individuals. What is
meaningful for one person does not have to be meaningful for somebody else. When we have a company that stores various types of statistical data, the data can serve financial analysis, accounting or marketing
strategy development. The data will be the same; the output may be
different. When we talk about storage systems, we are talking strictly
about data, processed in binary code, which consists only of ones and
zeros. A storage system cannot help us extract any kind of information.
Storage systems work only with raw blocks of data. Storage systems
themselves cannot see whether the data we store includes databases,
films, music files or other data types. Its important to remember that a
storage system sees only those ones and zeros.
28
Levels of data processing
Logical and physical layer
Distinction between data and information can also help us understand the more complex distinction between logical and physical levels of data. This distinction can be more difficult to grasp because
it is quite abstract. Generally speaking, we can say that the physical
level is the hardware level, with a primary focus on where exactly
raw blocks of data are stored on a hard drive, how they are optimized
for input and output (I/O) performance, and mutual communication
among hardware components on the lowest level. The logical level of
data, however, can be understood as a virtual interface working as a
bridge between the lowest level hardware operations and the user.
The user of the logical construct can be the server, the application or
the person using the equipment. This may sound confusing, but there
are many practical implementations of this abstract concept. The key
terms that will help you understand how physical and logical levels of
data work include physical unit, logical unit, partition, volume, file
system and microcode (to an extent).
In most of modern systems, the physical level of data processing is hidden from the user. The user either does not have the possibility to control
this level or can partially make changes using tools that are able to influence the physical level through the logical level.
Hard drive partitioning

Hard drive partitioning, well known from ordinary personal computers, can be a good way to explain the distinction between how logical
29
and physical levels actually work. In a model situation, we can have a

laptop running Microsoft Windows software. If you click on the My
Computer icon, you can access your file system where files are organized in a tree structure. On the top of this tree, you have icons of installed hard drives, CD/DVD drives and memory card drives. Each of
these drives is attributed a letter that serves as an address valid within
the operating system. A hard drive usually has C: and D: addresses. A
DVD drive usually has an E: address. Though rare in modern computers, a floppy disk drive usually has an A: address. You may notice that
on many computers with this type of configuration, you can have two
icons of hard drives (C: and D:) while you know that there is only one
physical drive installed inside. Why is this happening? Why is there not
only one icon representing the whole drive? It is because somebody,
perhaps you or a vendor, partitioned the disk into two logical disks. In
the computer, there is just one piece of hardware, but the user of this
hardware, which in this case is the operating system and you indirectly,
will see figure 1.6.
Boot Record C:
Master Boot Record

Physical disk drive
Logical Disk D:
Logical Disk C:
Boot Record D:
Figure 1.6 One physical disk drive is partitioned into two logical drives (C: and D:).
Now you may ask what the joining element between physical hard
drive and logical hard drives is. In this case, it is the metadata stored on
the first physical sector of a hard drive that contains the Master Boot
Record (MBR) table with information on the partitioning of the disks,
30
their proportions and their status (i.e., whether they are formatted and
contain some kind of file system). The MBR table is able to address only
disk drives with a capacity of up to 2TB. For bigger disks, there is a new
standard called a GUID partitioning table (GPT). This is, however, not
very important for us at the moment.
Why would people partition their hard drives? There can be several
reasons:
OS and application system files may be separated from user files
for purely organizational reasons.
It is easier to create a backup image of a partition that contains
the operating system. (It is not very practical to create an image
of the whole disk.)
You may want to install different file systems and operating systems on a single drive.
You may want to encrypt one partition that contains important
data.
Volume
Compared to the term partition, the term volume is more general
as it describes any single accessible storage area with asingle installed
file system. The term volume is mostly used in the context of operating
systems, but we can encounter this term even when we use software
for managing storage arrays. Volume will always be interpreted as a logical layer because it is not connected directly with physical operations.
When we have a physical hard drive with two or three partitions, the
first partition can contain a New Technology File System (NTFS) used by
the Windows NT series, the second partition can be formatted with the
ext3 file system used by Linux operating systems and the third partition
can remain unformatted, without a file system perhaps because we
31
have not yet decided what to do with this disk space. When we boot
Windows, we see only one volume, because Windows systems cannot
access and use the Linux ext3 file system (at least not without third party
utilities that may enable access to partitions that use different file systems). Windows also cannot use unformatted disk space. When we boot
Linux, we see two volumes one with the ext3 file system and one
with the NTFS, because Linux is able to access disk space that is using
the NTFS (even though Linux does not use the NTFS for its installation by
default settings). The third partition cannot be called volume because it
does not have a file system installed, and thus does not comply with the
definition of the term volume.
The term volume does not even have to be connected with partitions. If you insert a CD with music or a DVD with film, it will be again
called a volume. The same applies to USB pen drives, memory cards or
legacy floppy disks. They are not partitioned, and yet they are recognized as volumes by an operating system.
The term volume is then again part of the logical level of data and
it comprises the terms partition or disk drive and file system. The term
volume can be understood as a logical interface used by an operating
system to access data stored on the particular media while using a single instance of a file system. A volume, therefore, cannot contain two
or more file systems.
We have learned that you can split one physical hard drive into two or
more logical drives through a technique known as partitioning. If the logical volume contains a file system, an operating system sees it as a volume.
It is very important to note that a partition (or volume) can also be created from two or more physical hard drives. If you have two physical hard
drives and you want to create one unified volume, then you can join them
together, creating a single logical disc that will have the capacity of both
physical hard drives. This technique is called concatenation of hard drives.
32
File system
We have mentioned that volume is defined by two things it must
be a single storage area and it must have some sort of file system installed. We have encountered the term file system quite a lot throughout the last few paragraphs without describing what it actually is. In
simple words, a file system is a way in which files are named, organized
and stored on the hard drive. Again, it can be described as some sort of
logical interface between the user layer (individual files, for example,
.mp3 music files and .avi video files) and the physical layer that contains
only ones and zeros. File system translates raw binary data into what we
see when we open a file explorer in our preferred operating system. A
file system is, therefore, also a way of storing file attributes and metadata that contain additional information about the file and are used by
the operating system. This information can bear details such as access
permissions to users, a read only attribute, and the date of file creation
and modification. Please note that the file system is closely connected
with the users operating system choice. If you prefer to run Windows,
you must use NTFS. If you wish to run Linux, you must employ a different type of afile system, probably ext3.
To sum it up, a file system serves these purposes:
It sets up the way files are named. Applications such as text processors and video players use these names to access files. These
applications do not see where the particular file is physically
stored on the hard drive.
Applications access the file system using an application programming interface (API) and make requests, such as copy, paste, rename, create directory, update metadata and file removal.
A file system is always installed on a homogenous storage area
that is, in the context of operating systems, called a volume.
A file system stores information about where data is physically
located.
33
A file system stores metadata that contains additional information.

A file system maintains data integrity and allows you to set access
restrictions and permissions.
A file system can offer advanced functionalities, such as file compression, encryption and keeping track of various versions of a
single file (older and newer, using volume shadow copy technology).
There are more features of a file system, but the ones we discussed
are among the most important. Please note that storage systems do not
work with file systems, but with blocks. Blocks consist of raw data that can
be seen as a sequence of bytes or bits. Blocks can have various lengths,
meaning the number of bytes or bits that can be stored in one block depends on the blocks size. In a storage system, you have special hardware
processors that determine where to store the particular data depending
on a storage systems configuration and settings.
Figure 1.7 shows how a file system works in general terms. To have a
better idea how a particular file system works, we can take a closer look
at the NTFS used by Windows.
To understand how this particular system works, please refer to Figure 1.8, where you can see the Windows boot procedure. The MBR is
stored on the first sector of a hard drive. This record contains disk partitioning information, and also tells BIOS which partition is active and
bootable. The boot sector can be found on each partition and contains
information about the file structure (in Windows it is a tree structure).
The NT loader is a set of instructions describing how to use the file system and in which order to load all the drivers and libraries. At the end of
this process, we have a running operating system. Note the difference
between kernel mode and user mode. In kernel mode, the operating
system has direct access to installed hardware. In user mode, applications access these devices via the operating system and its drivers.
34
Picture.jpg
IInformation
f
ti
Picture
Operating
System
Worksheet1.
Holiday.png
Picture.jpg
user.dat
win.ini
DISK
DATA)
Figure 1.7 How a file system works.
Kernel Mode
User Mode
Hard Disk
Master
Boot Record
Boot
Sector
NT
Loader
Operating
System
Applications
Figure 1.8 How the NTFS works.
Logical units on a storage system

We are already familiar with conceptual distinction between physical and logical levels of data. Now it is time to reveal why this is so important when it comes to storage systems. In very broad terms, we can
see a storage system as a rather complicated and sophisticated bunch
of hard drives based on different technologies (according to requirements on performance). This bunch of hard drives creates a large storage area that is further partitioned into smaller storage areas that are
to be linked to individual servers. Usually, you have several servers with
running applications connected to one storage system. The partitions
35
you create on a storage system are basically logical units that are striped
across a large number of physical hard drives. When we talked about
partitioning, you learned that you can split one physical drive into several partitions. On a storage system, this works the other way around
you usually connect several physical hard drives, or their parts, to create one partition. This partition or logical unit is then assigned a unique
identification number, which is called a logical unit number (LUN). Remember this definition because we will be talking about LUNs many
times throughout this book. The LUN created on the storage system
physical hard drive is then linked to a server. This is called mapping.
We create a logical unit and then we map it to a server. When we map
a LUN to a server, the server then sees this LUN as a physical disc and
is able to perform read and write operations as if it really were a single
physical hard drive.
From this description, we can see that physical and logical layers
work on several levels. You have physical hard drives on a storage system and then you create a logical unit and map it to a server, which sees
it as if it were a physical hard drive. The server installs a file system on
this physical drive, making it logical to the user, which can be an application or an actual person sitting in front of the server.
Though this is already a very complex topic, it is actually just the beginning. Storage systems today create so many virtualized logical layers
that it is easy to get lost. At this point, it is also good to mention why
storage systems work on so many virtualized logical layers it is because of performance. In the next chapter, well explain how we achieve
performance through virtual logical layers.
36
To summarize, LUNs represent a logical abstraction or a virtualization

layer between the physical disk device/volume and the applications. Storage administrators can assign access and privileges on the basis of LUNs.
A logical unit that is created on a storage system over several physical
hard drives is mapped to a server using LUN identification.
In the context of storage systems, logical units are often referred to as
LUNs. The term LUN can, therefore, be used to describe just the unique
identification number or the logical unit itself.
Key words to remember are LUN and mapping.
Microcode
In terms of mere definition, microcode is built-in software that works
on the lowest layer of instructions, directly controlling hardware equipment. It is important to note that microcode is a term used not only with
computers and storage systems but with all electronic devices that are
based on microprocessor technology. As a cell phone or digital camera
user, you may have encountered the procedure of a firmware update
you connect your cell phone or digital camera to a computer and
it downloads new internal software (firmware) from the Internet. The
term firmware can be, in most cases, interchangeable with the term
microcode. However, firmware installed in your digital camera or even
in your washing machine usually contains microcode and some kind of
user interface (menu, icons, etc.), while the term microcode can be interpreted as the most basic software, with no graphical user interface
(GUI). Microcode is usually stored on a high performance memory chip,
which can be either read only or read and write.
37
In relation to storage systems, we can differentiate between two

types of microcode:
Microcode stored on individual hard drives: In storage systems, it
is common to update microcode stored on each individual hard
drive to fix bugs and increase performance of the whole system.
Microcode on a storage system that also contains an integrated
user interface, either a command line interface (CLI) or aGUI. In
this case, it is possible to rewrite or update the microcode. Some
storage system vendors use the term firmware instead of microcode.
Microcode, in a way, represents the interface between the physical hardware and the user, and can be represented by another piece of
hardware, application or even end user. It can also be seen as a set of
instructions that governs the relationship between the physical layer and
the logical layer.
Remember that when we talk about storage systems, the term microcode refers to the built-in software that contains basic tools for managing
disk arrays. To perform more complicated operations (LUN mapping, RAID
group setup, replication, etc.) you need additional software provided by a
storage system manufacturer. Storage systems do not require an operating
system to run; they use microcode instead.
Data consistency
Data consistency plays a crucial role in any data storing device. To
have consistent data basically means you have valid, usable and readable data. In other words, you need to have all ones and zeros in the
right order to be able to read data. If something goes wrong in the
storage device and just one bit of information is changed, then the
whole sequence of ones and zeros becomes unreadable and useless
38
for the end user, which can be an application running on the server. Therefore, it is vitally important to employ various technological
means to maintain data consistency and prevent data corruption.
When we use a computer at home, we usually have just one hard
drive installed. If this hard drive malfunctions and one sector becomes
unreadable, we have little protection and are likely to lose the whole
file stored on this bad sector. However, all modern hard drives have bad
sector detection tools included in their microcode, and when one particular sector on the disks shows signs of possible failure, the hard drive
automatically moves the data from this sector to a more secure place
on the platter. Another way to make sure your data remains consistent
is to back it up regularly to a data storing device.
The problem of data consistency is even more painful on large storage systems. Due to redundancy concepts, there is virtually no risk of
data inconsistency caused by a single hard drive malfunction; the problem is connected with ongoing data transactions. Imagine asituation
where a bank operates a server that runs an Internet banking application. This application stores data on an enterprise storage system.
The internet banking service is used by many users and there is a large
number of transactions made every second. If the server or the storage
system loses power, there would be at least one incomplete transaction
that hasnt been stored in time. When the bank gets the power back,
this incomplete transaction is likely to cause database inconsistency. It
can be very difficult to restore data consistency, and that is why storage
systems offer sophisticated software tools that in most cases are able to
achieve this result.
When we are talking about data consistency, we can differ among:
Point-in-time consistency
Transaction consistency
Application consistency
39
Point-in-time consistency means that data is consistently as it was

at any single instant in time. Point-in-time consistency is maintained,
for example, in synchronous (continuous) data replication, where data
is stored at the same time on two independent storage systems. At any
single point-in-time, the data must be the same on both storage systems.
To understand transaction consistency, you must first understand
the term transaction, which is a logical unit of work that may include
a number of files or database updates that occur in one shot. Transaction consistency is described in the example above the data will
not be transaction consistent if there was an in-flight transaction at the
time of device failure. In other words, if there was a transaction hanging somewhere in the air that was not processed before the shutdown
of the storage system, it would be lost. If this happens, the storage system is able to identify that there was a lost transaction and offers the
storage administrator various possibilities for fixing the problem.
Application consistency is similar to transaction consistency there
are just more streams of transactions belonging to a single application.
If we have an application that uses different kinds of databases and
files, we need to keep all streams of transactions consistent.
Data integrity
While data consistency deals with data readability and usability
based on the correct sequence of ones and zeros (if there is one mistake in the sequence, the whole file is corrupted and unreadable), data
integrity describes accuracy, reliability and correctness in terms of security and authorized access to the file. To put it in a different way to
maintain data integrity, we have to ensure data in a database is not
altered. If we lose data integrity, data is still readable and technically
usable, but it is not correct. If we disrupt data integrity, we replace one
40
sequence of ones and zeros with another sequence of ones and zeros
that also makes sense to the storage system (i.e., data is consistent) but
alters the information carried by the data in an undesirable way.
Data can be altered either by unauthorized user access or incorrect
application and it can endanger previously stored data by overwriting it
instead of, for example, keeping both copies of the file (the original and
the altered version).
It is important to maintain integrity and trustworthiness for the entire lifecycle of the data. The reason is that the differences among the
various versions of a single file or a single database entry can also carry
useful and, in some cases, essential information. To ensure data integrity, we have a large set of tools ranging from permissions and restrictions of access to a single file or a single database entry (we can set
the data to read only and forbid writing operations), and we can also
employ data integrity policies that consist of a set of rules and govern
access and possible data alterations.
Storage administrators must take data integrity into consideration
when designing data architecture, and they must be trained on how to
set rules governing data integrity. Data integrity becomes acomplicated
matter especially when it comes to structured data, or databases.
Thanks to cheaper and more powerful hardware, we have more and

more data coming into existence every day. This, together with increasing
dependency of people and businesses on quick access to data and information, brings an exponentially growing demand for storage.
Having read this chapter, you should be able to understand what kinds
of data exist and what their importance is to companies and individuals.
You should be able to name and describe the basic principles of data
management and terms connected with data storage systems. In later
chapters, we will discuss each of these terms in detail and you will see
what the technology behind these concepts is and how it works.
Storage components
andtechnologies

How a hard drive works
What means of hardware redundancy we have
How to describe midrange storage system architecture and
components
How to determine factors directly influencing the
performance of a storage system
How midrange and enterprise storage systems differ
42
2 Storage components and technologies
Overview of disk array components

In the first chapter, we discussed data and its importance. You learned
the basic principles, and now its time to get to something slightly more
technical well look at the storage system itself to see how these principles work when we implement them. By now you should know that
the very basic component of each storage system is a hard drive, a piece
of hardware well known from any personal computer. We know that
a hard drive is a device that stores data. Simply put, a storage system
is a highly sophisticated, state-of-the-art piece of technology that connects dozens or even hundreds of physical hard drives to create a disk
array. Other components of astorage system include disk array controllers, cache memories, disk enclosures, power supplies, fans and racks.
A disk array provides increased availability, resiliency and maintainability
by using additional, redundant components (controllers, power supplies,
hot-swappable spare hard drives, etc.). If one component of a storage
system fails it does not affect the functionality of other components and
it doesnt cause loss of data. We call this dual-controller architecture
or full redundancy; these concepts are implemented to have no single
point of failure. To help you understand how exactly this works, we will
start by taking a look at a hard drive and its construction. Then we will
describe how these individual hard drives are interconnected and how
this influences availability and reliability of data. Once youre familiar
with this concept, well move to other storage system components and
their descriptions.
How does a hard drive work?

A hard disk drive (HDD) is a nonvolatile storage device that stores
data on a magnetic disk. This is the basic definition. A nonvolatile storage device, unlike a volatile storage device, is able to retain data even
when the computer is not switched on when it is disconnected from
a power supply. Data is written on the disk by magnetizing particles
43
within a magnetic material in a pattern that represents the data. The

hard disk is able to read back this data by detecting the magnetic patterns created during the write process.
Key components of a disk drive are (see Figure 2.1):
Platter
Spindle
Actuator arm assembly
Read/write (R/W) head
Controller
The data is stored in binary form (ones and zeros) on the platters in
the shape of circular discs, which are coated on both sides with magnetic particles. The platter itself is made of nonmagnetic material, usually glass, aluminum or ceramic. The data can be read and written by an
actuator arm with an R/W head. Aspindle holds one or more platters
and is connected to a motor that spins the platters at a constant number of revolutions per minute (rpm). The set of platters spins in a sealed
case, called a head disk assembly. The velocity of the spinning platters
is one of the most important aspects that influences I/O data speed.
Common disks spin at 4,800rpm, 5,400rpm and 7,200rpm. The slower
drives are typically used in laptops because they consume less power
than the faster ones. For use in servers, we see hard drives that have
platters spinning at 7,200rpm, 10,000rpm or even 15,000rpm. These
hard drives are more expensive and their components must be more
durable to maintain a high level of reliability.
The space between the head and platter is so minute that even a dust
particle can disrupt the spin. HDDs are manufactured in clean rooms in
which air quality, temperature and humidity are tightly controlled. The
sealed cover also prevents intrusion of any dust.
44
platter
hard drive assembly (HDA)
spindle
actuator arm
actuator
read/write head
interface
power
connector
power connector
HDD mainboard with
controller
interface
Figure 2.1 A hard drive and its components.
The read/write head of an HDD reads data from and writes data to
the platters. It detects (when reading) and modifies (when writing) the
magnetization of the material immediately underneath it. Information
is written to the platter as it rotates at high speed past the selected
head. The data is, therefore, written in circles called tracks. The tracks
are then divided into sectors, which represent the smallest usable space
on the hard drive.
45
An actuator arm with an R/W head is another component that highly influences the performance of a hard drive. The actuator arm must
quickly find the data on the platters; this is called seek time. Average
seek times range from 8 10ms for common hard drives. However, in
the most modern hard drives designed for use in servers, it can be only
3 5ms.
Spindle
Platters
Figure 2.2 Spindle and platters.
As we already know, performance of an HDD can be measured by

velocity of platters in revolutions per minute and by seek time in milliseconds. There are two more basic parameters:
Capacity: The number of bytes an HDD can store. The current
maximum capacity of an HDD is 4TB.
Data transfer rate: The amount of digital data that can be moved
to or from the disk within a given time. It is dependent on the
performance of the HDD assembly and the bandwidth of the data
path. We can further differentiate between the internal transfer
rate from platters to cache (buffer), which is measured within the
HDD assembly, and the external transfer rate from the cache to
the bus. The average external transfer rates range from 50MB
300MB per second. See Figure 2.3.
46
Interface
HDD Bus
Storage
System
Cache
Hard Disk
Assembly
Controller
External transfer rate
Internal transfer rate
Figure 2.3 Internal and external transfer rate.
A hard drive is connected to other components of a computer or

storage system via an interface. The most common types of interface
are:
Parallel advanced technology attachment (PATA)
Serial advanced technology attachment (SATA)
Small computer systems interface (SCSI)
Serial attached SCSI (SAS)
At this point, it is important to emphasize that common hard drives
used in personal computers cannot usually be used in storage systems. Storage systems usually support only a few types of hard drives
(with strictly determined parameters such as capacity, rpm and interface). These hard drives are equipped with more electronic systems to
increase performance. The microcode of these drives contains highly
sophisticated monitoring tools that are able to determine the condition
of the hard drive and predict possible failures or malfunctions. Hard
drives to be used in disk arrays are usually supplied by the manufacturer of the particular storage system. They are sealed in a special plastic
drawer that is hot swappable. This hard drive component can be easily
plugged in and out of a storage system with no additional equipment,
such as a screwdriver.
47
Parallel advanced technology attachment)

PATA was a standard used to connect HDDs to computers, based on
parallel signaling technology. PATA cables are bulky and up to 18in long,
so they can be used only in internal drives, which is a serious limitation.
The PATA interface became obsolete for a few reasons. Firstly, the data
transfer rate is limited to 133MB/sec. Secondly, the hard drive connected via this interface is not hot-swappable, which means that if you want
to connect or disconnect a drive to the computer, you have to power it
down. Especially in the storage systems business, it is necessary that
all hard drives (and other components) are hot-swappable. When one
hard drive fails, you can replace it without rebooting the whole system.
Figure 2.4 A PATA cable.
Serial advanced technology attachment

SATA evolved from PATA. It uses serial signaling technology. SATA became a standard mainstream technology used to control and transfer
data from a server or storage appliance to a client application. Today,
SATA is also a standard in all personal computers. Compared to PATA,
SATA has the following advantages:
48
Greater bandwidth
Faster data transfer rates up to 600GB/sec
Easy to set up and route in smaller computers
Low power consumption
Easier manipulation
Hot-swap support
Figure 2.5 SATA cable.
Small computer systems interface

SCSI is a parallel interface standard used to transfer data between
devices on both internal and external computer buses. Compared to
PATA and SATA, SCSI can offer faster data speeds, multiple devices can
connect to a single port and there is a certain level of device independence. A SCSI interface works not only with hard drives, but also with
other hardware such as DVD-RW drives. There are some disadvantages,
though. SCSI interfaces do not always conform to industry standards,
and SCSI is more expensive than PATA and SATA. SCSI is rarely found in
personal computers, but it can be found in servers and storage systems.
SCSI has recently evolved into SAS.
49
Serial Attached SCSI

Serial Attached SCSI evolved from the previous SCSI standards; it
uses serial signaling technology. SAS is a standard used to control and
transfer data with SCSI commands from a server or storage appliance
to a client application. SAS can offer even faster data transfer rates than
SCSI. This standard also has the lowest power consumption. SAS cables
are similar to SATA cables. A SAS bus is compatible with older SATA hard
drives. It is thus possible to connect a SATA hard drive to a SAS bus. It is
not, however, possible to connect a SAS hard drive to a SATA bus.
SAS technology is very current and suitable for enterprise applications. It offers increased performance and inexpensive implementation (compared to Fibre Channel), and it natively supports redundancy
features such as dual bus for connection to two independent/interconnected controllers. Remember that SAS serves not only for connecting a
hard drive to a controller within an expansion unit, but also for connecting expansion units to a storage system back end interface.
To sum it up, PATA and SCSI interface technologies are becoming obsolete. SATA can be found in most modern personal computers. Both SATA
and SAS hard drives are suitable for use in servers and storage systems.
Solid-state drive (SSD) a hard drive with noplatters

A solid-state drive is a data storage device that uses solid-state
memory to store persistent data with the intention of providing access
in the same manner of a traditional block I/O hard disk drive. SSDs are
different from traditional HDDs, which, as discussed earlier, are electromechanical devices containing spinning disks and movable read/write
heads. SSDs, in contrast, use microchips that retain data in non-volatile
50
memory chips and contain no moving parts. Compared to HDDs, SSDs

offer a wide range of advantages. Due to the absence of moving parts,
they consume less power, thus producing less heat, which lowers demands on storage system cooling operations. These drives also provide
lower access time and latency. The major disadvantages are their price
and the limited number of writes over the life of the device. SSDs use
the same interface as HDDs, and thus easily replace them in most applications.
Connecting hard drives together

RAID levels and their function
You have learned how a hard drive looks and what its components
are. You have also learned that when you interconnect multiple hard
drives you get a disk array. Now its time to find out how this is done.
The key word is redundancy.
The technology behind connecting hard drives together is called redundant array of inexpensive (or independent) drives (RAID). The trick
is to combine several physical hard drives to create a single logical disk
that can be further split or partitioned into LUNs, as mentioned in the
previous chapter. The physical hard drives connected in one RAID group
must be of the same make and the same capacity they must be identical. The outcome of all this is that an operating system does not see
individual hard drives, but just one logical disk that combines the capacity of all individual hard drives.
51
RAID provides the following advantages:

Data consistency and integrity (security, protection from corruption)
Fault tolerance
Capacity
Reliability
Higher speed
andbetter
performance
Logical Disk
Physical Disk
block 1
block 2
Disk 1
Disk 2
block 1
block 5
block 2
block 6
block 3
block 7
block 4
block 8
block 3
block 4
block 5
block 6
block 7
block 8
Figure 2.6 A logical disk is acombination of physical disks.
Different types of RAID can be implemented, according to application requirements. The different types of RAID implementation are
known as RAID levels. Common RAID levels are: RAID-0, RAID-1, RAID1+0, RAID-3, RAID-5 and RAID-6. Each RAID level is a combination and
application of four features, or parameters, which are data striping,
data mirroring, data spanning and parity distribution. Selection and
implementation of RAID levels is closely connected with performance
requirements and amount of redundancy. See Figure 2.6
Data striping
Data striping is one of the basic techniques that help us achieve increased performance of a storage system. With one hard drive configuration, you store your files sequentially on the platter. When you wish
52
to access particular data, the R/W head on the actuator must find the
correct place on the spinning platters, and then it reads the sequence
of bits. This can be further complicated by data fragmentation, when
one file is stored on several locations within a single hard drive. This
is how data is accessed on personal computers. On a storage system,
performance of this sequential access to data is far from sufficient. Data
striping provides a solution to this problem.
Imagine that you have, for example, five hard drives and a single file.
You can divide this single file into five parts, and then you can store each
part on one of those five hard drives. When you want to access this file, it
will be read from five hard drives at once, combining their performance.
In a storage system, you concatenate these five drives into a single logical disk that can be further partitioned and mapped to a host (server).
Whenever you store something on this logical drive, a RAID controller
installed in a storage system will stripe the data and distribute it to the
physical drives in the fashion that will ensure the highest possible performance. Remember, by data striping, you can achieve increased read
and write speed, but you cannot improve seek time performance (with
the exception of solid state drives with no moving parts). The significant
disadvantage of data striping is the higher risk of failure when any of
the above mentioned five physical drives fails, you will not be able to
access your data anymore. In other words, there is no redundant data
that could be used for reconstruction of files stored on the damaged
hard drive. So data striping must be combined with other techniques to
ensure there is no data loss.
Data mirroring
Data mirroring is efficient and the easiest way to prevent data loss
due to malfunction or failure of one or more physical drives. The principle of data mirroring is quite simple read and write operations of
data are conducted simultaneously on two physical hard drives. One
53
file is automatically saved on two physical drives. This not only ensures
a high level of data redundancy, but it also provides increased read performance, because data is accessed on both drives. This performance
boost, however, is not as significant as in data striping. This is why these
two techniques data striping and data mirroring are implemented
together. Storage system data mirroring works on the hardware level in
the RAID controller. Note that because the data is written in duplicates,
the created logical disk is just half the size of the actual installed capacity of physical hard drives. The major disadvantage is costly application.
Data spanning
Data spanning, or linear data writing, is the simplest way of storing
data. It does not offer any increase in performance and no level of redundancy and it is therefore not suitable for enterprise application. It
can, however, be practical for home use or in small servers. The whole
trick is to concatenate several physical drives to create one logical disk.
The data is then written on this disk sequentially. In other words, when
one physical hard drive gets full, the data is stored on the next available physical drive. This can be useful when you want to have a unified logical disk on several hard drives with no demands on increased
performance. The main advantage of data spanning implementation is
that you do not need a hardware RAID controller. The other advantage
is that the system that consists of several physical hard drives is easily
scalable; when you run out of space, you can easily add more physical
hard drives. The last advantage is that if one physical hard drive fails,
the data on other hard drives is not lost [except for the file(s) that are
overlapping both physical hard drives]. Data spanning is also used on
CDs and DVDs. When you are making a backup image of your data, it
often happens that capacity of one backup medium is not enough. You
will therefore back up your system on several DVDs.
54
Parity
The last procedure for RAID implementations uses parity data to balance the disadvantages of data striping and data mirroring. Data striping
offers no redundancy and data mirroring is costly because you can only
use half of the physically installed capacity. If you want to overcome
these issues and get the best of all the above mentioned techniques,
you should use parity. In simple terms, parity data contains information that can be used to reconstruct and recalculate data on a failed
hard drive. Using parity requires at least three physical hard drives. You
would concatenate these drives to create a single logical disk. The logical disk offers twothirds of the physically installed capacity. Whenever
you perform write operation on this logical disk, the RAID controller
stripes data on two physical hard drives and then, using sophisticated algorithms, calculates parity data to be stored on the third physical drive.
Note that you can also stripe parity data. In this case, the RAID controller calculates the parity data and then stores data and parity stripes
on all three physical hard drives. Parity data works by complementing
actual data. The advantages are obvious you save money, and you
have a good level of redundancy and performance. The disadvantage is
that, if one disk fails, it takes time to reconstruct the data from the parity. However, the RAID controllers used in modern storage systems are
powerful enough to overcome even this issue, and when a disk fails, it
provides fast reconstruction times and prioritized immediate access to
required data.
Just a Bunch of Discs (JBOD)

This method of interconnecting hard drives cannot really be included in RAID levels since it offers neither redundancy nor better performance. In fact, JBOD is a term used for implemented data spanning.
Data is written and read sequentially. When one disk gets full, data is
stored on another disk in succession. Aside from the benefits we men-
55
tioned in the data spanning section, there is another advantage you

are not limited to using hard drives of the same make, capacity and
interface you can use whatever hard drive you have. JBOD is implemented in small servers or in computers that are designed to work with
large amounts of data, for example, computers for video editing.
RAID-0
In RAID-0, data is spread evenly across two or more disks, data is
stored just once without any backup copy. RAID-0 is easy to implement
and offers increased performance in terms of data access (more disks
equals more heads, which enables parallel access to more data records). This RAID level has no redundancy and no fault tolerance. If any
disk fails, data on the remaining disks cannot be retrieved, which is the
major disadvantage.
With a three disk implementation, one third of the
data is sequentially written to
each disk. In other words, if
you store a file to a logical disk
that consists of three physical
disks connected in a RAID-0
level, this file is then divided
into three pieces, and each
piece is stored on a separate
disk. We can see that RAID-0
uses only data striping.
RAID Level 0: Disk Striping

Block 1
Block 2
Block 3
Block 4
Block 5
Block 6
Block 1
Block 2
Block 3
Block 4
Block 5
Block 6
Block 7
Block 8
Block 9
Figure 2.7 Disk Striping.
56
RAID-1
RAID-1 implements mirroring to create exact copies of data on two
or more disks. In this way, it can offer increased availability of the data,
resistance to a disk failure and, thus, a good level of redundancy. RAID-1
is easy to implement, and it reduces the overhead of managing multiple
disks and tracking the data. It provides fast read/write speed because
the system can read from either disk. If a disk fails, RAID-1 ensures there
is an exact copy of the data on the second disk. This RAID level also has
some disadvantages. The most important is limited capacity in RAID1, the true storage capacity is only half of the actual capacity, as data is
written twice. This also makes RAID-1 implementation more expensive
it doubles the costs.
RAID-1+0:(2D+2D)
RAID 1
with Rotating Copy
A1
A1
A2
A2
A3
A3
A4
A4
8 Tracks
A1
B1
C1
D1
E1
F1
A2
B2
C2
D2
E2
F2
A1
B1
C1
D1
E1
F1
A2
B2
C2
D2
E2
F2
Disk 1
Disk 2
Disk 3
Disk 4
Primary Data
DISK 0
DISK 1
Primary Data
Figure 2.8 Diagram showing distribution of data

on hard drives set up in RAID-1 and RAID-1+0.
RAID-1+0
RAID-1+0 is an example of multiple, or nested, RAID levels. A nested
RAID level combines the features of multiple RAID levels. The sequence
in which they are implemented determines the naming of the nested
RAID level. For example, if RAID-0 is implemented before RAID-1, the
RAID level is called RAID-0+1. RAID-0+1 combine the features of RAID-0
and RAID-1 by mirroring a striped array. The advantages of this connec-
57
tion are obvious you get fast read/write speed and data protection
based on redundancy of data. The only disadvantage is high implementation cost.
RAID-5 and RAID-6

RAID-5 consists of a minimum of three disks (two data and one parity). RAID-5 distributes parity information across all disks to minimize
potential bottlenecks if one disk fails. If failure occurs, parity data from
the other disks is used to recreate the missing information. RAID-5 is
the most common and secure RAID level that balances performance,
security and cost. It ensures fast read speed and data recovery if a disk
failure occurs. The disadvantage is that extra overhead is required to
calculate and track parity data. It offers slower writes because it has to
calculate parity before writing data.
To implement RAID-6, at least four disks are required. This configuration is very similar to RAID-5, with an additional parity block, allowing
block level striping with two parity blocks. The advantages and disadvantages are the same as RAID-5, except the additional parity disk protects
the system against double-disk failure. This feature was implemented to
ensure the reliability of the SATA drives. The key value is that two parity
drives allow a customer to lose up to two HDDs in a RAID group without
losing data. RAID groups configured for RAID-6 are less likely to lose
data in the event of afailure. RAID-6 performs nearly as well as RAID-5
for similar usable capacity. RAID-6 also gives the customer options as to
when to rebuild the RAID group. In RAID-5, when an HDD is damaged,
the RAID group must be rebuilt immediately (since a second failure may
result in lost data). During a rebuild, applications using the volumes on
the damaged RAID group can expect severely diminished performance.
Acustomer using RAID-6 may elect to wait to rebuild until a more opportune time (night or weekend), when applications will not require
stringent performance. See Figure 2.9 for reference.
58
RAID-5:(3D + P)
Data is striped with parity

over RAID members.
D1
Parity
D3
D2
D2
D1
Parity
D3
D3
D2
D1
Parity
Parity
D3
D2
D1
Disk 1
Disk 2
Disk 3
Disk 4
Figure 2.9 Data distribution

across hard drives connected in
RAID-5 and RAID-6 groups.
RAID-6:(6D + 2P)
Data is striped with parity

over RAID members.
D1
D2
D3
D4
D5
D6
P1
P2
D2
D3
D4
D5
D6
P1
P2
D1
D3
D4
D5
D6
P1
P2
D1
D2
D4
D5
D6
P1
P2
D1
D2
D3
D5
D6
P1
P2
D1
D2
D3
D4
D6
P1
P2
D1
D2
D3
D4
D5
P1
P2
D1
D2
D3
D4
D5
D6
P2
D1
D2
D3
D4
D5
D6
P1
Disk 1
Disk 2
Disk 3
Disk 4
Disk 5
Disk 6
Disk 7
Disk 8
The last few pages provided you with fundamental redundancy concepts based on RAID implementation. You should also understand the direct connection of RAID implementation to the performance of a storage
system.
You now know what a hard drive is and how it works. You also know
basic principles of interconnecting hard drives together. In the next pages,
we will essentially build up a midrange storage system, taking a look at all
the components and their architecture.
Remember the key principles: data striping, data mirroring, redundancy and parity. You will hear about these often.
RAID level range

In the previous paragraphs, we described the most common RAID
levels in their minimal configuration. For example, we know that for
59
RAID-5 you need at least three hard drives two for data and one
for parity. However, this configuration can be extended. When we use
adistributed parity, we can, for example, have seven hard drives used
for data and one for parity and the result will still be RAID-5. Usually it
is more advantageous to use a higher number of hard drives than allowed by a minimal configuration, especially for performance reasons.
The more hard drives you have in a parity group, the more data stripes
created, which offers higher read speed. Storage systems from various
vendors support different RAID level configurations. These RAID level
configurations are also often called a RAID level range. The RAID parity
group is a building block of a RAID group. In other words, if you create a
RAID-5 group in an eight disk configuration (7D+1P) this will be your
parity group. Several parity groups will then make a RAID group. Remember that hard drives used in one RAID group must be of the same
make, capacity and interface.
Relative RAID performance

The table below shows the performance levels for different RAID
configurations relative to RAID-1+0 (2D+2P), which is assumed to be
100% for reads and writes. The illustration is not meant to be a recommendation or imply that one RAID group is better than another RAID
group, but to illustrate how performance of a RAID group should be
matched to the type of performance an application needs.
RAID Level
Random Read,
Sequential Read
Sequential Write
Random
Write
RAID-1+0 (2D+2P)
100 %
100 %
100 %
RAID-5 (3D+1P)
100 %
150 %
50 %
RAID-5 (7D+1P)
200 %
350 %
100 %
RAID-6 (6D+2P)
200 %
300 %
66.7 %
Note
Proportional to the
number of disks
Proportional to the
number of data disks
60
The second column shows that the random read and sequential read
performance is proportional to the number of disks, because the disks
can be accessed simultaneously. With sequential writes, there are no
reads involved as with random writes. Therefore, the performance is
proportional to the number of data disks.
As for the random writes column, the reason for the performance
difference between RAID-6 (6D+2P) and RAID-5 (7D+1P) is that RAID-6
(6D+2P) must process 1.5 times (see below) more disk I/Os than RAID-5
(7D+1P). Therefore, the random write performance in RAID-6 (6D+2P) is
33% lower than with RAID-5 (7D+1P). The number of disk I/Os in RAID-5
random writes is four (old data/old parity reads, new data/new parity
writes). The number of disk I/Os in RAID-6 random writes are six [old
data/old parity (P)/old parity (Q) reads, new data/new parity (P)/new
parity (Q) writes].
The values listed in the table were collected using the Hitachi Adaptable Modular Storage 2000 family system. The table is meant to illustrate that different applications may require different performance
characteristics from its RAID group.
Software and hardware RAID

To implement RAID, you need specific means, which can be either
software or hardware based. For purely software RAID implementation,
you need just an operating system and perhaps some third party application, which will simplify the procedure. The advantage is the cost
this solution is very cheap. The software based RAID implementation will, however, show lower performance than other solutions. Other
disadvantages include dependency on the operating system, slow and
problematic reconstruction of data if one hard drive fails and limitation to only a few basic RAID levels in their minimal configuration. Since
this solution is purely software based, there is no dedicated hardware
61
responsible for all the operations. All operations are controlled by the
central processing unit (CPU), which also calculates parity information.
Performance of the CPU is therefore diminished. As a result, this type of
solution is applicable only in home environments and is not suitable for
application in business environments.
Some manufacturers of mainboards for personal computers offers
dedicated RAID chip controllers. This is, in most cases, just marketing
because all the computing is done by the CPU. This type of solution
also does not usually support interfaces other than SATA or the now
obsolete IDE.
Hardware RAID controllers, on the other hand, offer much more than
their software counterparts. They include dedicated processing units
for parity data calculation, and they have cache memory for improved
I/O operations. If one disk fails, the reconstruction is automatic and
faster. You do not need any additional software, and you can configure
the RAID array and perform all the operations by accessing the built-in
microcode interface. Hardware RAID controllers can ensure data consistency by adding an independent backup battery that is connected to
cache and keeps it alive in case of power disruption. When the power is
back online, data in the cache is stored on the hard drive, and the user
does not lose any data. A hardware RAID controller is usually employed
in servers or in high performance professional computers. In storage
systems, you always have a powerful integrated RAID controller modified to support a large number of individual hard drives.
On personal computers, it is possible to combine software RAID with

hardware RAID. On the hardware level, you can create two 1+0 RAID
groups that can be further joined in RAID-0 on the software level.
In this way, you can stripe data across a striped and mirrored disk array,
gaining additional performance. This type of solution is possible, but not
implemented very often because of software RAID downsides. It is enough
to know it is theoretically possible.
62
Spare disks
In any storage system you must have spare hard drives so that if one
disk in a RAID group fails, there is another one ready to take over. The
damaged hard drive is later manually replaced with a new disk.
There are two methods that support sparing out of RAID group
data:
Correction copy
Dynamic sparing
Correction copy occurs when a drive in a RAID group fails and a compatible spare drive exists. Data is then reconstructed on the spare drive.
Figure 2.10 shows how the reconstruction is conducted depending on
the RAID level implemented.
Dynamic sparing occurs if the online verification process (built-in diagnostic) determines that the number of errors has exceeded the specified threshold of a disk in a RAID group. Data is then moved to the spare
disk, which is a much faster process than data reconstruction. See Figure 2.11 for details.
Notice that the speed of reconstruction of a failed disk depends on
the implemented RAID level. If we have a RAID-1 level, which uses only
mirroring and no parity information, the whole process of reconstruction is simple, because the system just copies the data from the mirrored drive to the spare disk. With RAID-5, the reconstruction will take
longer, because lost data must be calculated from the remaining disks
and parity information. In RAID-6, we have double-disk failure protection if two disks fail, the procedure takes even longer than in RAID-5.
In Figures 2.10 2.13 all the diagrams are marked with an I/O is possible title. This means the data is accessible even during the process of
reconstruction. Performance of storage will be affected, but you have
immediate access to your data because requested data reconstruction
is prioritized during the process. Remember that in the event of dou-
63
ble-disk failure on a RAID level other than RAID-6, you lose all the data
and you are not able to reconstruct it using conventional means. That is
why RAID technology is usually combined with other techniques to ensure redundancy and high availability. The most common of these techniques is data replication, which will be discussed in its own chapter.
RAID-5 (4D+1P)
I/O is possible
A, B, C,... : data
P: parity data
A, B, C,... : mirror data
Controller
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Spare Disk
RAID-6 (4D+2P)
I/O is possible
Controller
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
RAID-1
I/O is possible
C
Spare Disk
RAID-1+0 (2D+2D)
I/O is possible
Controller
Controller
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Spare Disk
Figure 2.10 This diagram shows the process of data reconstruction in event of failure of
one or more hard drives. In RAID-6 we have two parity disks: P and Q.
I/O is possible
The number of (correctable) errors has

exceeded the predetermined maximum
A, B, C,... : data
P: parity data
Controller
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Spare Disk
Figure 2.11 Dynamic sparing operations.
64
Sparing can be set in such a fashion that storage system controllers

check the health of the installed hard drives and anticipate possible future failure. Each hard drive has built-in diagnostics that monitor and
store various parameters, for example, Spin-Up Time, Start/Stop Count
and Seek Error Rate. When the possible future failure is detected, the
storage system automatically moves data to the spare disk. This saves
time, especially because copying data is always faster than recalculating
data from the parity.
Built-in diagnostic capability is part of all modern hard drives, even

those used in personal computers. This technology is called Self-Monitoring, Analysis and Reporting Technology (SMART). To access data collected
by a hard drive controller, you need special monitoring software able to
read and evaluate this data.
Note that diagnostics used in storage systems are more sophisticated
than SMART.
On most modern storage systems, correction copy can have two parameters set by a storage system administrator. These two parameters
are copy back and no copy back, and they influence the behavior of
the storage system once the failed drive is manually removed from the
storage system and replaced by a new hard drive. Depending on the
settings, the storage system can move data from the spare disk that was
used for data reconstruction to this new drive, or it can set this newly
added hard drive as a new spare disk while the former spare disk becomes the regular member of the RAID group.
In Figure 2.12, you can see the diagram of copy back settings.
Acopy back procedure takes less time than data reconstruction, but it
can still have a minor impact on the storage system performance. That
is why it should be conducted when the system is not on high load, for
example at night. Copy back parameter can also be used with dynamic
sparing technology.
65
Choice of either copy back or no copy back usually depends on the

storage system administrator alone. Since performance is not gained
from either of these parameters, the selected option mainly has organizational impact.
I/O is possible
After the failed disk is replaced
A, B, C,... : data
P: parity data
Controller
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Spare Disk
Figure 2.12 Correction copy with copy back parameter.
As you can see in Figure 2.13, if parameter no copy back is selected,

after a dynamic sparing process or correction copy, the spare disk becomes a permanent member of the RAID group and a copy back does
not occur. The replacement disk becomes the new spare disk.
I/O is possible
The dynamic sparing or correction copy
A, B, C,... : data
P: parity data
Controller
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Disk Drive
Spare Disk
I/O is possible
The failed disk is replaced
Controller
Disk Drive
Disk Drive
Spare DIsk
Disk Drive
Disk Drive
Disk Drive
Figure 2.13 Sparing out with no copy back parameter.
66
Storage system hard drive enclosures

A disk enclosure is another component of a storage system. In terms
of definition, we can find two possible interpretations:
The most widely accepted description is a chassis with a single
hard drive and additional electronics (e.g., interface conversion
board). A good example of this is an external hard drive, which is
basically a small box with a SATA drive and a chip that converts a
SATA interface to USB.
In relation to storage systems, a disk enclosure can be understood
as a hot-swappable hard drive unit. See Figure 2.14.
Figure 2.14 Hot-swappable hard drive units include a small drawer that simplifies disk installation.
In SATA hard drives, an additional board with electronics is included. These electronics turn the SATA
interface into a SAS interface with two ports for two
independent controllers to ensure redundancy. SAS
hard drives have this double architecture implemented natively.
For storage systems, the term hard drive enclosure often covers not
only the hot-swappable disk chassis, but the whole expansion unit that
is basically a rack for installing many individual physical hard drives. In
67
other words, in the context of storage systems, the terms hard drive
enclosure and expansion unit are interchangeable.
An expansion unit or disk enclosure (based on SAS architecture),
consists of following components:
Individual hard drives (SAS, SATA, SSD)
Expander (buses and wires for connecting the drives together)
Power supplies
Cooling system (fans)
Chassis
Power/Ready/Warning/
Alarm/HDD LEDs
Disk drives (15)
Front view
Rear view
Power units
Figure 2.15 Front and rear view of an expansion unit/disk enclosure, Hitachi Adaptable Modular Storage 2000 family.
In Figure 2.15, you can see how a typical expansion unit looks. Note
that we usually have two types of expansion units a standard expansion unit, which can usually accommodate 15 hard disks, and a high
density expansion unit, which can accommodate up to 50 hard drives.
In high density expansion units, the disks are installed vertically and are
not directly accessible from the front panel. The advantage is that you
save space. The disadvantage is more complicated access to individual
disks and demand for more powerful cooling system (more disks generate more heat). Also notice the status LED lights that are a standard
68
part of regular expansion units and serve for monitoring purposes. Another reason each disk is accompanied with LED lights is localization
of a particular hard drive. Imagine that one disk in the enclosure fails.
The reconstruction process starts immediately and a spare disk is used.
An illuminated LED marks the failed disk that needs to be manually replaced. In high density expansion units, this is more complicated since
you need to pull out the whole expansion unit from the rack, remove its
cover and then localize the failed hard disk.
On the rear side of an expansion unit, you can see ports for connecting the enclosure to the back end controller of a storage system. These
ports can be based either on Fibre Channel or SAS technology, whichever is more efficient. We will describe the operations and differences
of these two technologies in detail later in this chapter. You can also see
power cord outlets.
In midrange storage systems, the first enclosure is usually part of the

base unit, together with other components. In some storage systems, the
base unit enclosure uses the first few hard drives for storing system administration software. The remaining capacity of these hard drives cannot
be used for user data.
We know that each SAS disk is equipped with two SAS connectors.
An expansion unit has two expanders each one is connected to one
SAS connector on each hard drive. If we use a SATA hard drive, we need
an additional device that will convert one SATA port into two SAS ports.
All this is done to meet cause of redundancy requirements. Storage systems are also equipped with double architecture every component
in a storage system is duplicated to maintain the no single point of failure requirement. This means that we also have two back end controllers that are connecting expansion units. Any hard drive is, therefore,
connected to both back end controllers via expanders and SAS cables.
See Figure 2.16.
69
Base unit
CTL0
CTL1
Expansion
unit
Expansion
unit
Expansion
unit
Expansion
unit
Expansion
unit
Expansion
unit
Expansion
unit
Expansion
unit
Figure 2.16 Each expansion unit is connected to both controllers.
Back end architecture

A midrange storage system contains main controller boards that are
equipped with components that can be put into three categories:
Front end (connection to hosts or other storage systems)
Cache (works as a buffer, has major influence on performance)
Back end (connection to hard drive enclosures, RAID operations)
Since we are building a midrange storage system step by step from
hard drives to complex solutions, the back end is another component.
Please refer to Figure 2.17 to see how the individual components are
connected.
70
When we are talking about back end architecture, we should know

that there are two basic ways of interconnecting individual disks in
enclosures. One is based on Fibre Channel technology and uses Arbitrated Loop topology. This architecture is, however, becoming obsolete,
and that is why we will pay more attention to SAS technology, which is
becoming a standard in back end architecture. More information about
Fibre Channel Arbitrated Loop can be found later in this chapter.
Individual components of SAS back end architecture are depicted
in Figure 2.17:
DCTL/RAID Disk controller with RAID processor. Determines
where to store the particular data coming from host via front end
and cache. Takes care of striping and calculates parity data. In
event of disk failure, it will control data reconstruction.
SAS controller Together with SAS buses, it provides the interface for connecting enclosures with individual hard drives.
WL/cable SAS wide link; a type of SAS cable.
Expander Serves for connecting individual hard drives.
AAMux Multiplexer that changes SATA interface to SAS interface (SAS encapsulation).
CTL0 and CTL1 Names of the controllers; notice double architecture.
Cache High-speed memory; a key component that stands between back end and front end.
71
Base unit
CTL0
Front end
DCTL/RAID
Front end
cache
cache
SAS
Controller
WL
WL
Expander
WL
SAS
Bus
DCTL/RAID
SAS
Controller
SAS
AAMux
SATA
SAS
Bus
cable
CTL1
WL
WL
Expander
WL
SAS
Bus
SAS
Bus
cable
Expansion unit
ENC0
ENC1
WL
Expander
WL
SAS
AAMux
SATA
WL
Expander
WL
Figure 2.17 Back end SAS architecture scheme.
Fibre Channel Arbitrated Loop (FC-AL) architecture

To understand operations and advantages of SAS back end architecture, we need to become familiar with back end architecture based on
the Fibre Channel Arbitrated Loop. Even though this solution is becoming obsolete and is not used in new models of storage systems any-
72
more, it is still present in installations that run applications on older

equipment. Note that SAS back end architecture is quite new introduced in 2008.
The name Fibre Channel Arbitrated Loop is derived from the three
basic parameters that define this type of architecture:
Fibre Channel Data is transferred through fiber optic cables in
the form of light impulses. These cables are made from glass or
plastic.
Loop All the hard drives are sequentially connected to a cable
structure that resembles a loop. A loop connection is an alternative to a point-to-point connection.
Arbitrated Only one hard drive can communicate with the
controller at a time. On the loop, the token is circulating. A token,
which is essentially a bit of data, determines which hard drive
can communicate with the controller. If you need to write data
on a particular disk, you need to wait until the token reaches this
disk, because only then is the communication possible. In other
words, only the hard drive that has the token can communicate
with the controller. When the I/O operation is finished, the token
moves to another hard drive. Since data is usually not read or
written in a sequential manner, you have a longer response time
because you have to wait for the token to reach its destination
hard drive. This is the element of arbitration. See Figure 2.18.
In FC-AL technology, you have a single link for one path that supports
a 4Gb/sec transfer speed. To meet redundancy requirements, each hard
drive is connected to two loops, each one belonging to one controller
(CTL0 and CTL1).
73
TOKEN
PATH 1
FC Port
Back end
Controller
5
6
Figure 2.18 Fibre Channel Arbitrated Loop back end architecture. Only one hard drive can
communicate with the controller at a time. A token determines which device can communicate
with the controller; other hard drives have to wait for their turn.
Do not confuse Fibre Channel Arbitrated Loop, which is becoming

obsolete, with Fibre Channel itself. Fibre Channel can deliver speeds up
to 12.75Gb/sec and is suitable for application over distance, since Fibre
Channel cables can be up to 100km long. A Fibre Channel connection is
often used between servers and storage system (front end). It is also the
building block of Storage Area Networks (SANs) that represent the storage
system networking infrastructure.
SAS back end architecture in detail

In comparison with FC-AL, SAS can provide a point-to-point connection, where hard drives communicate directly with controllers. The controller does not have to wait until the token reaches the destination
hard drive because, in SAS back end, there is no such thing. In the con-
74
figuration with four SAS links, there can be four hard drives communicating with the controller at the same time, each with a data transfer
rate of 3Gb/sec. We know that FC-AL can deliver 4Gb/sec. In SAS architecture we can, therefore, reach three times higher performance. FC-AL
provides good performance with larger chunks of data written sequentially, while SAS performs much better with random I/O operations. We
can illustrate this with an example: imagine that you have a train and
a car. Train cargo transportation is efficient when you have lots of load
directed to one destination. On the other hand, the car will be more
efficient with small loads to multiple locations. Since storage systems
must handle more small random I/O operations, SAS back end architecture offers better performance.
Controller 1
SAS wide cable
Expander
Controller 2
SAS wide cable
Enclosure
15 Disks
4-SAS
Links
4-SAS
Links
SAS
Disk
Slot 0
SAS
Disk
Slot 1
SAS
Disk
Slot 2
SATA
Disk
Slot 3
SAS
Disk
Slot 4
4-SAS
Links
SATA
Disk
Slot 5
SSD
Disk
Slot 13
To
next
tray
SSD
Disk
Slot 14
4-SAS
Links
To
next
tray
Expander
Figure 2.19 SAS back end architecture. Hard drives are connected to expanders, which are
placed in enclosures. In this picture, we can see an example of a four SAS link expander.
SAS based back end solutions provide other advantages in addition

to those we have already mentioned. SAS hardware is cheaper than FCAL hardware, thus it is more powerful technology for less money. SAS is
also able to handle failures more efficiently than FC-AL because it uses
component level reporting. When a hard drive fails, this particular drive
is mapped out and the failure is reported by asystem monitoring tool.
FC-AL, on the other hand, does not report at the component level, but it
reports any failures at the loop level. The outcome is that it takes longer
to correct the malfunction. The SAS solution, therefore, boasts higher
75
availability and allows for greater redundancy.

SAS also natively supports full duplex. Full duplex communication
means that both devices (controller and hard drive) can talk to each
other at the same time. In other words, both devices are able to transmit and receive I/O simultaneously. A half duplex solution is similar to
communication over walkie talkies only one side can talk at a time.
Fibre Channel technology delivers half duplex or full duplex depending
on the system architecture.
Cache
We have described individual hard drives in detail and how they are
connected together. You learned that the hard drive is the component
that stores data, and its performance is measured in transfer data rate.
In paragraphs dedicated to RAID, you learned that data striping can increase performance of a storage system the storage system performs
more I/O operations per second. I/O operations are basically read and
write requests coming from the host, which is the server. Nevertheless, hard drives are still the major bottleneck in storage systems. Their
performance is not sufficient, simply because of the limitations of their
construction. When you want to perform I/O operations with a hard
drive, you have to wait for the platter to turn into the right position and
for the actuator to move the head on the spot where you want to write
or read data. You may object that solid-state drives already overcame
these limitations, but they are still very expensive and their limitation is
somewhere else solid-state drives offer a reduced number of writes
per one memory unit, compared to common hard drives with platters.
The technology of solid-state drives is advanceing, but it still needs perfection. The component that drastically improves storage system performance is cache.
76
In storage systems, cache is volatile high-speed memory that can

temporarily store data. Volatile means that this type of memory is
erased when the power is turned off. Lets compare it with the performance of a hard drive hard drives connected in RAID with data striping and cache:
A standalone hard drive can deliver external transfer data rates
ranging from 100MB/sec 300MB/sec. If we take 150MB/sec as
the average value, the average seek time is 5ms.
RAID-5 in 3D+1P configuration increases data transfer rates to
approximately 450MB/sec for read operations (data is read from
three drives). The write speed also increases, but not as drastically, since there is time required to calculate parity data and this
data also must be written (on four drives). You still have 5ms seek
time because of the construction limitations of a hard drive.
Cache can offer speeds up to 8GB/sec (8,000MB/sec) with access
time under 1ms. You can, therefore, see that cache speed is significantly higher.
When a host server makes write requests to the connected storage
system, it has to wait at least 5ms for the platters and actuator to turn
to the right position, plus time for the data transport itself. Another
read or write operation cannot be performed until the ongoing one is
marked as completed. Instead, it must be queued, which slows down
the whole system. Furthermore, you do not usually have a single server connected to a single storage system you can have hundreds of
them. Response time of a storage system without cache would not be
sufficient. Response time is defined as the duration of the cycle when
a server makes a request for a write operation and the storage system
sends confirmation that this write operation has been successfully completed.
The reason cache improves performance of a storage system is based
on the assumption that some of the data stored in a storage system
is accessed more often than other data. This frequently accessed data
77
can, therefore, be temporarily stored in high performance memory that

is expensive and can have limited capacity compared to non-volatile
storage devices. This function of a storage system is especially useful
for read operations. For write operations, cache operates more as a
buffer. With cache implementation, whenever the server makes a write
request, data is written to cache and then to the installed hard drives.
The storage system, however, will send the confirmation of the successful write as soon as the data is stored in cache. The storage system
waits to collect more write requests to one destination, thus making the
write operations more efficient since several write commands are sent
in one batch instead of sending several write requests to the disks. The
transfer of data from cache to the physical drives is solely dependent
on a storage system. The host gets the information of a successful write
when data is stored in cache, considers the whole operation completed
and immediately sends another request. Thats why cache dramatically
reduces astorage systems response time.
Two kinds of cache can be implemented: global cache for both read
and write operations or dedicated cache that supports either read or write
operations, but not both. Modern systems are equipped with global cache.
In midrange storage systems, the cache provides capacity of at least
8GB 16GB. Cache is the component that lies between the storage systems front end and back end.
Note, that individual hard drives also have small built-in cache to improve their performance. Capacity of this built-in cache ranges from 8MB
64MB. This cache basically works on the same principles as cache in a
storage system. Built-in cache improves performance of the external data
transfer rate compared to the internal data transfer rate.
78
Cache operations
Cache operations are controlled by the CPU and interface board that
are usually part of the front end. Because the capacity of cache is limited, efficient data management is of crucial importance. To maintain uninterrupted and fast data flow from the host to the back end hard drives
and vice-versa, the storage systems microcode contains algorithms that
should anticipate what data is advantageous to keep (for faster read access) and what data should be erased. There are two basic algorithms
that affect the way cache is freed up:
Least Recently Used (LRU) Data that is stored in cache and has
not been accessed for a given period of time (i.e., data that is not
accessed frequently enough) is erased.
Most Recently Used (MRU) Data accessed most recently is
erased. This is based on the assumption that recently used data
may not be requested for a while.
The capacity of cache is shared by two basic kinds of data read
data, coming from the hard drives to the servers, and write data, coming from the servers to the hard drives. The ratio of cache available
for read and write operations is dynamically adjusted according to the
workload.
Other algorithms govern the behavior of cache when it is getting
full or is underused. When the cache approaches the value set by the
microcode, dirty data is immediately sent to the hard drives for write.
Dirty data is a term for write data that has been confirmed to the server
as written but has not been written to actual hard drives yet.
Other operations are based on configuration. Storage systems are
usually supplied with software for cache management and cache partitioning. Cache partitioning can be useful because various applications
running on servers have various requirements on cache. Optimal configuration of cache highly influences performance of the whole storage
system and the whole infrastructure.
79
Cache data protection

Cache is volatile memory dependent on a constant power supply,
and it contains write data that appears to be written to the server, but
is in fact still in cache and not yet on the hard drives. This all means
that cache is an extremely sensitive and potentially vulnerable component that requires careful handling. Since the storage systems are based
on double architecture that provides no single point of failure, cache
must also be equipped with the means that ensure redundancy. The
first technique of data protection in cache is cache mirroring. Both independent controllers (CTL0 and CTL1) have their own cache module.
Both these modules are of the same size and technical specifications.
Thanks to cache mirroring, the cache modules on both controllers contain the same data. This brings protection in the event of a controller
or cache module failure. However, it does not solve the dependency on
electricity.
To overcome dependency on constant power supply, we have batteries installed in a storage system. In case of power failure, all components of a storage system shut down immediately. Installed batteries
only keep data in cache alive. Depending on the capacity of installed
batteries, data can be kept in cache up to 36hrs.
Some storage system vendors equip their devices with a cache vaulting functionality. The storage system that supports cache vaulting must
be equipped with more powerful batteries. In the event of power failure, batteries support not only data in cache, but also other components of the base unit. The only subsystems that shut down immediately are connected expansion units. Data from cache is then dumped
on dedicated hard drives called vault drives. When the data is safely
placed on the hard drives, the whole system shuts down. When the
power is back online, cache is restored using the cache image stored on
the vault drive.
80
Front end architecture

When we talk about the storage system front end, we usually mean
the interface board with individual ports that connect the storage system to the SAN with servers. The main component, therefore, is the
interface board. The interface board bears controllers for the particular
port. Common front end ports are:
Fibre Channel (FC) The medium is optical fiber wire that supports high speeds. Fibre channel is usually the best option you
can go with when building a SAN.
Fibre Channel over Ethernet (FCoE) The medium is metallic
(copper) wire, but the network protocol is still based on Fibre
Channel. This solution is cheaper but cannot offer as high atransfer speed as Fibre Channel. There are also limitations in copper
wire length. This port is becoming obsolete and in modern storage system it is replaced by the iSCSI port.
iSCSI Interface that connects the storage system to the LAN.
The advantage is that customers do not have to invest money in
building a SAN, and can instead use their network infrastructure.
This solution is suitable for smaller companies that use only one
or a few midrange storage systems. The medium is always metallic (copper) wire.
Aside from the interface board, the front end includes the main CPU
and random access memory (RAM) used for calculating and conducting
instructions connected with storage system operation. The main CPU
and RAM complete the instructions coming from the microcode and administration software, which serve for RAID group creation, LUN mapping, LUN resizing, replication, etc.
The front end interface board and CPU are also responsible for managing paths. In relation to storage systems, the term path describes
which way the stream of data goes from an application to individual
LUNs. For redundancy purposes, we usually need two paths from a
81
server to one LUN. One of these paths goes via CTL0, and the other goes
via CTL1. Similarly, back end enclosures (with individual hard drives) are
connected to back end controllers.
(4) 8Gb/sec Fibre Channel Ports
QE8
Controller 0
Celeron
1.67GHz
1 GB RAM
MCH
+ICH
QE8
PCI-e x8
2GB/sec
PCI-e x8
(4) 8Gb/sec Fibre Channel Ports
QE8
Hitachi
Adaptable
Modular
Storage 2100
QE8
Controller 1
Celeron
1.67GHz
MCH
+ICH
PCI-e x8
2GB/sec
PCI-e x8
1 GB RAM
2 x 2GB/sec
RAID Processor
(DCTL)
4GB/sec
2-4GB
Cache
2-4GB
Cache
4GB/sec
PCI-e x8
SAS CTL
Expander
SAS Wide Cable

(2 ports per controller)
4 links 3Gbit/sec each
4-15 HDDs
PCI-e x8
RAID Processor
(DCTL)
Internal
Enclosure
SAS CTL
Expander
1
SAS Wide Cable

(2 ports per controller)
4 links 3Gbit/sec each
Tray 0
15 HDDs
internal
Tray 1
Tray 2
15 HDDs
15 HDDs
Tray 3
Tray 4
24 HDDs
ENCLOSURE
STACK
24 HDDs
Tray 5
High Density
Tray
120-159 Disks
(SAS, SATA, SSD)
Figure 2.20 Overview of the whole mainstream midrange storage system architecture and
components using Hitachi Adaptable Modular Storage 2100 as an example. In the front end,
we have QE8 Fibre Channel port controllers that are part of the interface board. These controllers are mainly responsible for conversion of FC transfer protocol into PCIe bus used for internal
interconnection of all components. The storage system in this configuration is not equipped
with FCoE or iSCSI ports because they are optional. Notice the CPU and local RAM memory (not
cache). This processor is the microcode engine or the I/O management brain.
82
LUN ownership
Earlier we discussed that, in a storage system, we have to configure
data paths. The data path starts in a server that has two FC ports for
connection to two FC ports on a storage system one port is located
on CTL0 the other one is on CTL1. From the interface board, data is sent
to cache. The DCTL/RAID processor then takes care of data striping and
sending the stripes to their destination physical hard drives.
On the user level, we have to configure where the path leads we
have to map the LUN to a server. From one server to one LUN, we have
two paths for redundancy. When one path fails, data can be transferred
through the second path. As you learned, in a storage system we have
double architecture with two controllers for redundancy, but so far, we
have not mentioned that double architecture also equals double performance if no malfunction occurs. Since the storage system is accessed
from several servers, there are also several paths leading to different
LUN destinations. Imagine you have five servers accessing five LUN destinations. This makes 10 data paths. To use all hardware means efficiently, each controller is assigned a certain number of LUNs. When we create LUNs, we must also set LUN ownership, or decide which controller
will be responsible for which LUN. This way we can split the workload
efficiently between two controllers.
Data will then be transferred primarily through the path that leads
to the controller that owns the particular LUN. The second path is used
as a backup. In event of failure of the path that leads to owning controller (lets say CTL0), the backup path is used. However, the backup path
leads to CTL1, which is not set as a LUN owner. Therefore, data is transferred internally to CTL0, bypassing the malfunctioning FC port.
The path failure is depicted in Figure 2.21 on the next page. This configuration, when one controller is dominant and the second one is for
backup purposes, is called active-passive architecture.
83
Host
1. The host changes access path
because of a path failure
Port
CTL 0
Port
CTL 1
2. LU ownership
is not changed
LU0
Figure 2.21 Active-passive architecture. In the event of a path failure, LUN ownership does not
change. Data is transferred via the backup path to CTL1 and then internally to CTL0, bypassing
CTL0 front end ports. This diminishes performance because of internal communication.
Active-passive architecture means that only one connection (path) to

adevice can be active at a time. Other paths are passive with respect to
that device and waiting to take over if needed.
In an active-passive storage system, if there are multiple paths to one
LUN, one of them is designated as the primary route to the device. The
other one is designated for backup purposes only.
Modern storage systems, however, add the functionality of activeactive architecture, which brings several major advantages. In activeactive architecture, you can use more than one path to a single LUN
simultaneously.
84
HBA HBA
HBA HBA
HBA HBA
HBA HBA
LUN:0
LUN:1
LUN:2
LUN:3
CTL0
CTL1
Figure 2.22 Multiple paths to a single LUN are possible. LUN ownership automatically changes
in the event of path failure. Unlike active-passive architecture, active-active architecture offers
equal access to the particular LUN via both paths. This means the performance is not influenced
by what path is currently used.
If both paths are working correctly, both of them can be used for
communication. In Figure 2.22, you can see that one path from the first
server to LUN:0 is connected directly, and the second path to the same
LUN leads via CTL1 and also include internal communication. This requires communication overhead; however, this overhead has been reduced drastically through the implementation of modern PCIe buses in
modern storage systems. The PCIe interface allows fast internal communication between CTL0 and CTL1.
In the event of path failure in active-active architecture, LUN ownership can be dynamically changed. When the storage administrator
creates the LUN, ownership to the controller is assigned automatically.
85
In the event of path failure, LUN ownership is automatically switched

to another controller, and the particular LUN is mapped directly via this
controller. This mapping is more performance efficient since internal
communication performance overhead is eliminated completely and
the path goes straight through the controller. Active-active architecture enables other techniques that allow performance increase, such
as controller load balancing.
Users do not select a controller
when a LU is being created.
If necessary, LU ownership can be
changed manually afterwards.
Administrator
Firmware
- Create LU0
- Create LU1
- Create LU2
- Create LU3
LU0
LU1
LU2
LU3
Figure 2.23 Activeactive architecture.

No need to configure
LUN ownership manually. Communication
goes either via CTL0
or CTL1. In the event
of path failure, no bypassing as depicted in
Figure 2.21 is necessary; therefore there
is no communication
overhead.
The active-active front end design of modern storage system controllers allows for simultaneous host access of any LUN on any host port on either controller with very little added overhead. A host accessing a LUN via
a port on CTL0 can have most of the I/O request processed completely by
CTL1 with little intervention by the main CPU and DCTL processors in CTL0.
86
Controller load balancing

Simultaneous use of multiple paths to a single LUN together with LUN
ownership dynamic allocation are two major features of active-active
front end architecture. When employed together, they can eliminate bottlenecks and balance workload on both controllers. Imagine a situation
where CTL0 is 75% busy while CTL1 is only 15% busy. To reach optimal
usage of hardware means the load balancing monitoring tool, which is
part of the microcode, can change the LUN ownership to the less busy
CTL1, which will be used for data processing. You could object that both
paths could be used simultaneously, so LUN ownership is not necessary,
and simple usage of both paths would solve the problem. This is not true,
though. You need to realize that the DCTL processor is the exact location
where the bottleneck is created. The DCTL processor must calculate all
the parity data, and it rules where the data is to be distributed in terms of
physical hard drives. Remember that the LUN is still only virtualized space
within one RAID group. To avoid overloading a single DCTL processor, the
LUN ownership is changed and the less busy controller takes over. Figure
2.24 depicts how is this all done. Read the description of all the components carefully to avoid getting lost in the complexity of the diagram.
SW
Port
SW
Port
Bottleneck
Load balance
monitoring
MPU
MPU
CTL 0
CTL 1
LU0
LU3
LU1
LU2
SW
Port
Bottleneck
recovered
SW
Port
MPU
MPU
CTL 0
CTL 1
Automatically
LU0
LU3
LU1
LU2
LU1
Before
After
Figure 2.24 Controller load balancing diagram. Black boxes at the top of the pictures are host
servers. SW stands for switches, which are part of the storage area network (discussed in the
next chapter). MPU stands for microprocessing units (DCTL together with MPU). Ifone MPU is
too busy while the other one is underused, LUN:1 ownership is automatically changed to CTL1
and the corresponding path is redirected.
87
At this point, it is good to mention that one more element plays arole
in using several paths simultaneously. If you have one host server connected to a single LUN via two paths, then the servers OS will see two
mapped LUNs because it has no chance to realize that there is only one
LUN connected via two paths. Thats why there must be an additional
software layer installed on the host server. This software layer is based
on dynamic link provisioning and enables the server to work with the
mapped LUN via two paths. Note that dynamic link provisioning cannot
get the information that one particular controller is too busy. That is
why the storage system must redirect the path to aless busy controller, exactly as in Figure 2.24. Notice that the green arrow in the After
diagram is redirected within the storage system, even though it would
be better to change LUN ownership and communicate directly via the
second path. This example demonstrates that even active-active architecture is still imperfect.
Fibre Channel ports and their configuration

A midrange storage system usually offers at least four Fibre Channel
ports on both controllers that are used for connection to other devices
in network. These devices can be:
Host servers, usually connected via Fibre Channel switches. LUNs
are mapped to servers so that applications running on these servers can use the allocated capacity.
Another midrange storage system used for data replication. On
the connected storage system there are pair LUNs. In data replication, the source LUN is marked as P-VOL (primary volume) and
destination LUN is marked as S-VOL (secondary volume that contains exact replica of data on P-VOL).
A storage system with virtualization capabilities. This is usually
an enterprise storage system able to virtualize space on the midrange system and use it as extended capacity. This is especially
88
useful if you have several storage systems from various vendors

and you need to consolidate your infrastructure to lower costs.
Virtualized capacity is then managed through the enterprise system software tools.
Depending on the type of midrange storage system and vendor, Fibre
Channel ports can be configured as either an initiator or target. Some
modern midrange storage systems do not require any settings at all and
automatically configure the Fibre Channel ports according to the connected device. In enterprise storage systems, you usually have to configure ports manually. Setting the Fibre Channel port as the target serves
for connecting host servers or systems with virtualization capabilities. In
the case of astorage system replication pair, the P-VOL initiates the communication when replicating data to an S-VOL. On the source storage
system, therefore, we set the Fibre Channel port as initiator, and on the
destination storage system, we set the Fibre Channel port as the target.
iSCSI interface
iSCSI interface is an alternative to Fibre Channel interface. Fibre Channel ports are powerful, efficient and can be used for transmitting data
over long distances. However, their disadvantage is that they require the
entire network infrastructure to be built on Fibre Channel technology,
using Fibre Channel optical wires. This means that you have to build a
new SAN, and all the components, such as switches, servers and other
storage systems, must support Fibre Channel. This can be costly. To balance output performance and costs, it is possible to use iSCSI front end
ports that are often optional parts of a midrange storage system. The
first letter in iSCSI means Internet. This technology is therefore built on
a combination of TCP/IP and SCSI protocol. Through the technique of
encapsulation of SCSI instructions to TCP/IT protocols, data can be transmitted without special requirements on a network environment (i.e., a
specialized SAN network). This allows you to connect your storage sys-
89
tem to a LAN or WAN (especially Internet) using common Ethernet cabling. The advantage is that you can use your infrastructure and you do
not have to invest in costly Fibre Channel components. The downside
is that the performance of this solution is much lower. Therefore, iSCSI
connections are usually implemented in small companies that use only a
few midrange storage systems. Any storage system equipped with iSCSI
ports also offers Fibre Channel ports, so it is not complicated to upgrade
to an Fibre Channel infrastructure if needed.
Remember that a midrange storage system consists of two controller

boards, CTL0 and CTL1, which contain all the components. Components,
according to their tasks and roles, can be attributed to the front end or
back end. Between front end components and back end components
there is cache, which has a major impact on the performance of the whole
system because it offers high processing speeds compared to hard drives.
Enterprise storage systems

So far we have been discussing midrange storage systems. Some
vendors call their midrange products modular storage. Midrange or
modular systems are well suitable for application in small or medium
sized businesses and organizations. They are powerful in processing I/O
requests from a limited number of servers. One application accessing a
storage system makes one thread that needs to be processed. For companies that have a large number of servers running many applications,
enterprise storage systems are more suitable since they deliver significantly higher performance when it comes to multi-thread processing.
In other words, enterprise high-end storage systems are able to process
a large number of I/O requests simultaneously. Note that enterprise
storage systems can also be called monoliths. That is because enterprise systems are installed in proprietary racks of non-standard dimensions, which are supplied from the particular storage system vendor.
90
Enterprise storage systems offer a higher level of reliability, scalability and availability. We know that midrange storage systems have two
independent controller boards, which contain components that can be
attributed to front end, cache or back end. In enterprise storage systems, these components are installed on dedicated boards that can be
replaced individually. They are also often hot swappable, so there is no
need to turn off or reboot a storage system when replacing. In enterprise storage systems, you would have one board for the DCTL controller, one board for cache, boards for individual front end ports, etc. This
also means that you can install more than one instance of the component, which makes enterprise storage systems more flexible and more
scalable. In case you decide to upgrade your enterprise storage system,
for example, with additional ports or cache, you just need to order the
specific board that is then installed into the grid switch that provides
the platform and interface for mutual communication of the installed
components. For redundancy reasons, whichever component you order is always duplicated to maintain no single point of failure. The set
of two identical components that are to be installed in the grid switch is
called an option or feature. You can, therefore, order one cache option
that will consist of two identical modules, one for CTL0 and one for CTL1.
Enterprise storage systems can offer additional performance by
providing:
A large storage capacity more back end ports can be installed
and more expansion units can be connected to these back end
ports.
A significantly larger amount of cache to service host I/O optimally.
The ability to deliver high performance in multi-thread processing it can handle much higher number of I/O requests from
multiple hosts.
Higher throughput the internal buses and interface are able to
handle more data faster than midrange storage system. Throughput is measured in GB/sec.
91
Fault tolerance architecture to improve data availability there

are more means for hardware and software tools to provide undisrupted access to data.
Connectivity to mainframe computers and open system hosts.
Often, the ability to manage other storage systems via the process of storage space virtualization to consolidate a heterogeneous infrastructure.
Support of automated dynamic tiering.
Sophisticated support of data migration.
Mainframe computers are the most powerful computer systems that
have evolved from the first powerful computers created in late 1950s.
These computers used to occupy entire rooms and even now consist of
several racks. Mainframe computers are extremely expensive and extremely powerful. They are used largely in government institutions and
research facilities, such as NASA.
Storage space virtualization of a heterogeneous infrastructure
means that you can use the enterprise storage system for controlling
other storage systems (modular and enterprise) and creating a single
unified virtualized storage area that can be further divided into logical
units. This is especially useful for companies that have storage systems
from different vendors because they must pay for software licenses,
maintenance and employee training for each system, which becomes
costly. With one virtualized storage area, companies can manage all
their storage systems from different vendors with one software tool,
and pay only for one kind of license software, maintenance and employee training on a single product.
Automated dynamic tiering is a storage system functionality that
automatically moves data to hard drives of different performance according to frequency of access. This is closely related to data lifecycle
and archiving. For now, it is important to know that in storage systems
terminology, a tier is defined as a storage area based on asingle tech-
92
nology. Therefore, you can have a fast tier for frequently accessed data
based on SSD drives. Then you can have a tier based on SAS or SATA
drives or an archiving tier with tape backup technology. Automated dynamic tiering is very useful and will be discussed later when we talk
about archiving.
In Figure 2.25, we can see the Hitachi Virtual Storage Platform, which
is an enterprise storage system that brings state-of-the-art technology
and many innovative features. One of them is 3D scaling, which allows
you to expand the storage system into three dimensions you can add
performance and storage space, and you can connect other vendors
storage systems and virtualize them.
Figure 2.25 Hitachi

Virtual Storage Platform
is an example of an enterprise storage system. In
this picture, we have two
VSP systems with expansion units installed in additional racks.
This chapter provided you with detailed information on the design and
architecture of midrange storage systems. Now you should be able to:
Name all individual components and explain their function.
Explain basic concepts of achieving and improving performance.
Explain basic concepts that ensure data redundancy.
In addition, we compared midrange and enterprise storage systems.
Both are based on very similar concepts and architectures the main differences are in performance and application.
Storage networking
and security
How to describe basic networking concepts
How common network devices operate
What possibilities we have in storage system networking
How devices communicate with each other through
anetwork
How to secure storage area networks
94
3 Storage networking and security
Introduction to networks
Nowadays, a standalone piece of IT equipment is of little use. It is
the ability to communicate with other devices that makes technology
so powerful. People would hardly spend so much time in front of their
computers if they could not share the outcome of their work with other
people. When we say the ability to communicate with other devices is
important, we especially mean the ability to transmit and receive information encoded in the form of data to and from other members of
a network. A network, therefore, describes an infrastructure of technical means that facilitate this interchange of information. These means
are both hardware and software based. Among the hardware network
components, there are cables, network adaptors installed in personal
computers, servers, storage systems or other devices, and components
dedicated for routing the stream of data to its destination. The software layer of networking consists mainly of protocols that contain sets
of instructions that govern the process of communication among the
network members. Each protocol supports a specific area of application (i.e., type of network) and type of network adaptor. For each type
of network, we need adedicated solution through a particular set of
cables, network adaptors and protocols. Storage systems are highly sophisticated devices that must be able to provide extremely high performance. As a result, storage systems require a different type of networking than regular workgroup computers. In this chapter, we will focus
primarily on storage system networking.
Why do we need networks?

There are several reasons we use networks. Before we get to the
technical descriptions of individual component designs and functions,
it is good to clear up what motivation people have to use a network.
A network interconnects electronic devices to establish communication. People can use many tools to communicate: email, instant messaging, telephone and video.
95
A network enables people to share information in the form of

data. It is possible to access files stored on the other side of globe
within seconds.
Thanks to networking, we can share technological resources.
For example, one office printer can be used by multiple users.
This feature is also very important when we talk about cloud solutions, when end users access and take advantage of remote
computing means. This means we can have common computers
connected to a cloud computing service that does all the calculations. In this way, the end user with a common computer can
use applications that require performance beyond possibilities of
personal equipment. We can also employ distributed computing
that allows us to merge the performance of several network devices to complete the task.
This is the motivation behind network implementation. But networks also have several inherited properties that need to be taken into
consideration. The major threat of networking is its potential security
risks. Each network can be the subject of cyber attacks and unauthorized access. Computer viruses and worms use networks to spread.
Therefore, networks must be secured. To secure a network, we have
a range of hardware and software means, such as firewalls and antivirus programs. A network infrastructure may also interfere with other
technologies. This applies especially for wireless networks where data is
transmitted over air. Networks are also very complex in terms of implementation, configuration and maintenance. Careful infrastructure planning is important because networking can be costly and require highly
professional personnel.
Network cables
Network cables are the basic building block of any wired network.
They serve as the medium through which data flows from one network
96
device to another. The type of cable chosen for a network is closely

related to the networks type, topology, protocol and size. Each cable
type has its characteristics that influence network performance and determine the area of application. The two basic kinds of cables used in
networking are twisted pair and fiber optic.
Figure 3.1 Twisted pair cable structure, RJ45 connectors.
In Figure 3.1, you can see an example of a twisted pair cable. Atwisted pair cable is usually made of copper and has pairs of insulated wires
twisted together. The twisting reduces electromagnetic interference
and cross-talk from neighboring wires. Twisted pair is the least expensive type of cabling used in networks. However, it provides limited distance connectivity. This means that you cannot use twisted pair cables
in a wide area network (WAN). The maximum available throughput
depends on the cable quality and construction. The tighter the twisting, the higher supported transmission rate. The cables allowing higher
speeds are also more expensive. The common throughput ranges from
10Mb/sec to 1Gb/sec. The maximum cable length reaches only 100m.
Twisted pair cable usually use RJ45 standardized connectors well known
from common network adaptors installed in personal computers and
laptops.
Cable Jacket
Strengthening Fibres
Coating
Cladding
Core
Figure 3.2 Fibre Optic Cable.
97
Fiber optics is a technology that uses glass or plastic fibers to transmit data as light impulses. A fabric optic cable consists of a bundle of
fibers; each fiber can transmit millions of messages modulated onto
light waves. Usage of glass or plastic materials makes the cable resistant
to electromagnetic interference. However, these materials also make
the cable fragile. That is why additional cladding and strengthening fibers are added. Fiber optic cables have the following advantages over
twisted pair cables:
Provide a greater bandwidth
Allow data to be transmitted digitally
Can carry data over long distances
All these advantages make fiber optic cables ideal for WANs. The
main disadvantage of fiber optic cables is that they are expensive to install. The decision to use a twisted pair or a fiber optic cable in a network
depends on the protocol used. It is possible to distinguish between multi-mode optical fiber and single-mode optical fiber. Multi-mode optical
fiber is suitable for communication over short distances, up to 600m.
This points to the area of application in an organizations infrastructure.
Single-mode optical fiber is suitable for long-distance communication.
The difference in technology is that multi-mode optical fiber is cheaper,
since it has a larger and more robust core that allows cheaper implementation of light emitting diodes (LEDs) and vertical-cavity surfaceemitting lasers (VCSELs). In simple terms, the manipulation with multi-mode optical fiber is easier and less expensive to implement. Fiber
optical cables are color coded to easily show the type. Single-mode
optical fiber cables are yellow, and multi-mode optical fiber cables
are orange. However, purely for practical reasons, sometimes we use
different colored cables of a single type to distinguish among paths in a
network and simplify cable management.
As for the connectors that are used with fiber optic cables, the situation is more complex than with twisted pair. There are many standards
of fiber optic connectors depending on the area of application. Optic ca-
98
bles are used not only in networking, but also for transmission of audio
signal and other purposes. The common connector for fiber optic cables used with networking of storage systems is called LC, which stands
for Lucent connector (name of manufacturer), little connector or local
connector. Fiber optic cable connection to aport also requires a special
device able to convert optical signals into electrical signals and vice
versa. This device is called a transceiver. Sometimes the transceiver is
built-in as part of a port, but more often it is a standalone component.
There are two types of transceivers commonly used in storage system
networking:
Gigabit interface converter (GBIC)
Small form-factor pluggable (SFP)
Figure 3.3 Fiber optic cable LC connector and SFP transceiver.
It is possible to say that both GBIC and SFP do the same work, while
the GBIC solution is older and becoming obsolete. SFP transceivers are
smaller and offer higher speeds. Both types of transceivers are highly
reliable and hot swappable. There are several sub-types of SFP transceivers that are implemented according to the area of application.
For multi-mode fiber with orange color coded cable coating:
SX 850nm, maximum 4.25Gb/sec over 150m
99
For single-mode fiber with yellow color coded cable coating:

LX 1310nm, distance up to 10km
BX 1490nm, two paired cables for uplink and downlink (BS-U
and BS-D respectively) for distances up to 10km
EX 1550nm for distances up to 120km
In storage systems networking you will be mostly dealing with orange

color coded multi-mode optical fiber cables with LC connectors and SFP
SX transceivers that are marked as 850nm and support 4.25Gb/sec and
a150m distance.
Network members nodes

A node transmits or receives data over a network. Nodes are therefore any devices that participate in mutual communication among all
network members. A node can operate as a communication end point:
personal computer, tablet, server, storage system. Anode can also operate as a redistribution point, when the device redirects astream of
data from the communication originator to the communication target.
Each node has its unique identification, which serves for addressing
purposes in the network. We will look at the most important means of
network identification.
An Internet Protocol (IP) address is a numerical identifier assigned
to each device connected to a computer network that uses Internet
Protocol for communication. Addressing based on IP identifiers is common in local area networks (LANs). In other words, you can encounter
an IP address at your home network. An IP address looks something like
this: 192.168.1.1.
100
A media access control (MAC) address is a worldwide unique identifier assigned to a network interface card (NIC). This identifier is not
dependent on protocol, such as IP addressing, because a MAC address
is built into the device by its manufacturer. This is why MAC addresses
are sometimes referred to as physical addresses. A MAC address looks
something like this: 01:2F:45:67:AB:5E. As you can see, hexadecimal
digits separated by colons or hyphens are used. Behind hexadecimal
digits there is a 48b binary code.
A World Wide Name (WWN) is a unique label, which identifies
aparticular device in a Fibre Channel network. To be more precise, it
is used for addressing Fibre Channel communication targets or serial
attached SCSI (SAS) targets. As you may guess, WWN addresses are
used in storage area networks (SANs), the primary focus of this chapter. Later we will discuss the structure and usage of WWN addressing.
Except for such nodes as computers, printers or even storage systems, which we call endpoint network members, we have network devices that serve primarily as data stream redistribution points. To understand how data is transmitted and distributed, it is necessary to get
familiar with the Open Systems Interconnection (OSI) model definition
of network communication layers.
The lowest communication layer, the physical layer, is represented
by astream of bits and their transport over a network. On this level of
communication we deal just with ones and zeros.
The second layer is called the data link layer, in which data frames
are transported over a network. A frame consists of three basic elements: a header, payload field and trailer. A payload field contains the
usable data we want to transfer the data that conveys information.
The header and the trailer carry metadata that notify the beginning of
a frame, the end of a frame and, most importantly, the originating device and target device physical addresses (i.e., MAC or WWN). The primary function of the second OSI layer is, therefore, physical addressing.
Thanks to this metadata, we know exactly where the frame is coming
101
from and where it is heading. Another defining feature of a frame is

that it has fixed size, i.e., the number of bits dedicated to the header,
payload field and trailer. The size of a frame depends on the type of
network and protocol used.
The third layer, which is the last one thats important for us at the
moment, is the network layer. On this level, we are dealing with packet
transfer. The packet determining feature is the use of logical addressing
to allow network routing. In network devices, packets are encapsulated
into frames that are then sent over a network.
data unit
Host Layers
data
data
data
Media Layers
segments
packets
frames
bits
layers
application
Network Process to Application
presentation
Data Representation & Encryption
session
Interhost Communication
transport
End-to-End Connections
and Reliability
network
Path Determination &

Logical Addressing (IP)
data link
Physical Addressing (MAC & LLC)
physical
Media Signal and Binary Transmission
Figure 3.4 The OSI model.
Communication over a network is a very complex process that works

on several layers represented by the OSI model. The highest layer is the
application level. When an application wants to communicate over a network, it passes data to a lower layer of processing, e.g., an operating system, network protocols and drivers or network interface cards, while the
lowest level is binary code. The layer system is implemented because of
work division an application itself cannot work directly with frames,
and frames alone are of no use to an application. The process of passing
data to a lower layer is called encapsulation. The process of encapsulation
can be understood as putting a sheet of paper into envelope. The sheet of
paper contains data, the envelope address, a stamp, and so on.
102
When we talk about network data stream redistribution nodes, we

mean particularly network hubs, switches and routers. Each of these
devices has its own area of application. For different types of networks
and network cables, we need different types of hubs, switches and
routers.
Hubs a hub is a simple device that allows interconnection
and communication among nodes. A network hub works on the
first, physical, OSI level. It accepts the stream of data from one
port and simply distributes it to all other ports, which increases
network traffic and reduces network performance. Ahub therefore cannot direct the stream of data to the particular destination and it does not provide any network security features.
Network hubs have no intelligence and almost no possibilities
for configuration. A hub is inexpensive but not scalable.
Switches a switch also allows interconnection and communication among nodes within one network. Unlike hubs, network
switches are working on the second layer of OSI model. They use
physical addressing and they work with frames. When a frame arrives at the switch, the switch reads its header, retrieves the MAC
or WWN identification number and then sends the frame directly
to its destination. Switches are therefore much more efficient in
network communication than hubs. Switches are possible to configure and can provide network security features, such as zoning
(discussed later).
Routers a router provides an interface between two different
networks, for example, between a WAN and LAN. A router works
on the third layer of the OSI model and uses logical addressing.
Logical addressing is represented by IP addresses. A router receives and transmits packets according to their IP addresses.
Routers can be used at home to share Internet connectivity. A
router is assigned an IP address for communication with the providers gateway. The router then assigns local IP addresses to all
the computers connected in the LAN, enabling them to access
the Internet. Routers are configurable.
103
Except for these network redistribution points, you can encounter

other devices such as repeaters, which amplify the network signal, or
bridges, which can enhance the effectiveness of network communications over a LAN.
Client
Server
Nodes
Hub
2
Storage
Printer
3.5 Network nodes are all the devices connected in the network. We distinguish between
endpoint communication nodes and data redistribution nodes.
Storage networking topologies

As we already know, storage system networking is based on Fibre
Channel technology. In Fibre Channel, we have several possibilities of
interconnecting the individual nodes. The way in which are nodes connected to each other is commonly called network topology. The main
Fibre Channel topologies are:
Point-to-point or FC-P2P
Arbitrated loop or FC-AL
Switched fabric or FC-SW
104
A point-to-point topology is considered the simplest topology, in

which two devices are directly connected using Fibre Channel. It has
fixed bandwidth, and data is transmitted serially over a single cable.
This topology can be implemented as direct attached storage (DAS).
Point-to-point topology supports legacy products and can be used for
tape backups.
Arbitrated Loop (AL) is an FC topology we already know from storage
system back end architecture. FC-AL networking works on very similar
principles. All devices are part of a loop, and only one device can communicate with another device at a time. In AL, devices use an access
request mechanism called a token, or arbitrate (ARB), which circles
the loop. A device can use ARB depending on its priority and access
rights; the device with the highest priority gets first access. Arbitrated
loop topology is designed for mass storage devices and other peripheral
devices that require very high bandwidth. Arbitrated Loop is usually deployed in small environments and in tape drives because it is inexpensive and easy to implement, and it supports legacy products. The main
disadvantage is that adding and removing a device from the loop interrupts all activity on that loop.
The hardware that connects workstations and servers to storage devices in a SAN is referred to as a fabric. The SAN fabric enables any-server-to-any-storage device connectivity through the use of Fibre Channel
switches or network directors. In switched fabric topology, bandwidth
is not shared between devices, enabling devices to transmit and receive
data at full speed at all times. Switched fabric is, therefore, the most
powerful and complex topology and it is also the preferred Fibre Channel network topology implementation. It supports multiple concurrent
point-to-point connections; it is a highly reliable, available and scalable
solution. The main disadvantage of switched fabric topology is cost it
requires at least one switch to establish connections between devices,
ideally two for redundancy.
105
2
1
Switch
Fiber Chanel
Storage
ge
Figure 3.6
Switched fabric
topology. All the
devices in the network can communicate with each
other any time at
full speed.
Direct attached storage (DAS)

Storage systems offer high performance. To take advantage of this
performance and inherited redundancy features, storage systems require specific networking solutions. The easiest way to connect a storage system into a network infrastructure is through direct attached storage. In direct attached storage, a storage device is directly connected to
an individual server equipped with a host bus adaptor card (Fibre Channel interface). A DAS device is not shared, so no other network device
can access the data without first accessing the server.
DAS is mainly in a configuration where primary data access is required by the applications that run on the directly connected server.
The advantages of this solution are lower initial costs. You do not need
Fibre Channel switches to build a DAS network. The concept of DAS also
does not mean that you can connect only one server to the particular
storage system. Any storage system is equipped with more than one
front end Fibre Channel port. In midrange storage systems it is usually
four front end Fibre Channel ports. This effectively means that you can
connect up to four servers in DAS. In the event that you would want to
maintain no single point of failure, you would use two paths from the
server connected to two front end ports in a storage system. It is also
possible to connect clustered servers.
106
The disadvantages are that one or two servers usually do not use
the full potential of a storage system, which is achieved by sharing the
hardware means across the network.
LAN
Server
Storage
Figure 3.7 Direct attached storage infrastructure. Server is directly connected to a storage system. Storage system can be accessed only through the server. Server can be accessed from LAN.
Storage area network (SAN)

To meet high performance requirements and use all possibilities
storage systems offer, we can create network infrastructure dedicated
to storage systems. Because this network type is dedicated to implementation only with storage systems, it has features tailored to ensure
a high level of redundancy, fault tolerance and high availability. A storage area network is a high-speed network of shared storage devices.
Servers attached to a SAN can access any SAN attached storage devices.
The components in a SAN are storage devices and switches. A SAN is
designed to connect computer storage devices, such as disk array controllers and tape libraries, to multiple servers or hosts.
In a SAN, the processing power of a production server is reserved
for applications. Therefore, a SAN implementation ensures faster processing of the applications running on this server (i.e., there is no overhead). In DAS, the processing performance of the attached servers can
be diminished because it must provide part of its resources for requests
from hosts connected in the LAN. Therefore a communication overhead
exists due to accessing the storage system through the server.
107
In addition, SAN has the following advantages:

Effective disaster recovery
Redundancy and fault tolerance
High-speed access
High security
The main function of a SAN is to transfer data between computer
systems and storage elements, such as storage devices and switches.
A SAN moves all data storage to a separate network, used only for
data. ASAN is ideal for complex data centers, multiple data centers or
when applications and departments need access to shared data. A SAN
meets requirements on high availability, scalability, reliability and performance.
Disaster recovery is a formal plan that describes how an organization

will deal with potential disasters, including precautions to maintain or
quickly resume mission-critical functions.
Fault tolerance is the ability of a system to continue to perform its
function at a reduced performance level, when one or more of its components have failed. Fault tolerance in disk subsystems includes redundant
instances of components whose failure would make the system inoperable, where redundant components assume the function of the failed ones.
High availability is the ability of a system to perform its function continuously for a significantly longer period than the reliabilities of its individual components would allow.
Scalability is the ability to grow or support growth so all capabilities of
the system remain balanced. For example, a storage system whose data
transfer capacity increases by the addition of channels as its storage capacity increases by the addition of disks is said to scale.
Reliability is an attribute of any computer component (software, hardware or network) that consistently performs according to its specifications.
108
Storage area network components

Storage area networks use Fibre Channel infrastructure, which includes host bus adaptors installed in servers, Fibre Channel cables and
switches, and Fibre Channel ports installed in the storage system front
end and proprietary network protocols.
WWN
48-bit format
12-bit extension
4-bit prefix
64-bit
WWN example - 5 0 0 6 0 E 8 0 1 0 4 05 3 0 3 0 1 6
Figure 3.8 Host bus adaptor and an example of WWN.
In a storage system, an HBA is a Fibre Channel interface card installed

in a server. It connects a computer and storage devices on anetwork.
HBAs can offer one or two Fibre Channel ports for connecting optical
cables via SFP transceivers. Each HBA has a unique WWN, which serves
as a physical address, similar to MAC addresses used in LAN interface
cards. There are two types of WWNs on an HBA:
Node WWN: Shared by all ports on an HBA
Port WWN: Unique to each port on the HBA
As you should already know, a World Wide Number is a unique builtin identification used to address each element on a Fibre Channel storage network. Each WWN is a 64b number derived from the Institute of
Electrical and Electronics Engineers (IEEE) standard. The 64b number
includes a 48b format, a 12b extension and a 4b prefix.
Fibre Channel switches are different from those used in LAN. They
are used in small to midsized SANs. They allow the creation of a Fibre
Channel fabric, which is the shortened name of Fibre Channel switched
fabric topology (FC-SW). In FC-SW, devices are connected to each other
through one or more Fibre Channel switches. This topology allows the
109
connection of up to the theoretical maximum of 16 million devices, limited only by the available address space (224). Multiple switches in a
fabric usually form a fully connected mesh network, with devices being
on the edges of the mesh.
Fibre Channel switches are not highly reliable or scalable. To ensure
high availability and no single point of failure, each node needs duplicate paths across the SAN via independent Fibre Channel switches.
A Fibre Channel director is a large and complex switch designed for
application in large storage area networks such as enterprises or data
centers. It is highly available, reliable, scalable and manageable. It is designed with redundant hardware components, which provide the ability
to recover from a non-fatal error. Fibre Channel directors are capable of
supporting mainframe connection via FICON protocol. FICON protocol
will be described in detail later in this chapter.
Storage area network over iSCSI interface

SAN can be implemented by using a network protocol standard called
iSCSI, which uses the SCSI protocol to transmit data over TCP/IP networks. iSCSI allows organizations to utilize their existing TCP/IP network
infrastructure without investing in expensive Fibre Channel switches.
iSCSI uses TCP/IP as an alternative to Fibre Channel to overcome the
limitations of Fibre Channel, such as high expense and distance limitation. iSCSI has the following advantages:
Facilitates data transfers over intranets
Manages storage over long distances
Is inexpensive to implement
Provides high availability
Is robust and reliable
110
iSCSI is best suited for web server, email and departmental applications. iSCSI ports can be found in a storage system front end interface
board. Storage systems usually support both Fibre Channel ports and
iSCSI ports.
Network attached storage (NAS)

Network attached storage is a highly specific solution that differs
from common SAN storage systems in many ways. The main difference
between storage systems and NAS systems is that NAS can be defined
as a file-level data storage device that provides file access over a network to heterogeneous clients. We know that SAN storage systems are
operating on the data block level and that the file system layer is installed in connected servers. A SAN storage system therefore needs a
server with running applications that can use a mapped LUN capacity
that appears to the server as a physical hard drive. Compared to this,
NAS is very different.
NAS operates not on the block level but on the file level. This means
that a NAS device has a file system installed. A NAS device does not
provide LUNs that need to be mapped it provides network locations
where it is possible to store and read data in the form of files. A NAS
device is attached directly to LAN infrastructure enabling all hosts to
access the files stored in NAS. The host computer cannot see volumes
as physical installed hard drives, because there is no LUN mapping.
The host computer sees the network location volumes. These volumes
can be accessed from any device in the LAN. A NAS device can be a
specialized storage system that supports NAS functionality (NAS appliance), a server equipped with one or more hard drives connected in
RAID, a server with a NAS application connected to a midrange storage
system, or an enterprise storage system equipped additionally with a
NAS option (or NAS blade). Large and powerful NAS storage systems
allow virtually thousands of client connections to access a single file
111
simultaneously. See Figure 3.9 to get a general idea of how NAS solutions actually look.
Clients
NAS Head
Application
Server
Print
Server
Storage
NAS Device
Figure 3.9 Network attached storage. The NAS device is represented by the server that functions as the NAS head and common storage system. There are solutions that integrate both
these functionalities in one package (NAS appliances). NAS devices work relatively independently; they do not require servers with applications. All clients, application and other servers
can access files stored in a NAS device.
A NAS device is sometimes called a filer because the focus of its

processing power is file services and file storage. Supported file access protocols are Network File Protocol (NFS), Common Internet File
Protocol (CIFS) and protocols from the Internet environment (HTTP,
secured HTTPS and FTP). NAS devices support Fibre Channel and iSCSI
connections between the NAS head and a storage system. The NAS
head is always connected to a LAN using standard RJ45 connectors. A
NAS system moves the overhead of file system processing and serving
off the host to a dedicated device or devices optimized for these tasks.
NAS devices are made up of the following components:
112
Network interface using one or more interface cards (NIC), typically 10Gb Ethernet
Network file system protocols (NFS, CIFS)
Internal or external RAID storage
File system
Cache
Industry standard storage protocols to connect to and manage
physical disk storage resources
The advantages of a NAS solution are that it offers storage to computers running different operating systems over a LAN. It can be implemented within the common network infrastructure with no requirements for SAN Fibre Channel technology. It solves the problem of
isolating data behind servers typical for DAS and even SAN block-level
storage systems. NAS devices are optimized for sharing files between
many users. Implementation to current infrastructure is easy.
The disadvantages are that a NAS solution relies on the client-server
model for communication and data transport, which creates network
overhead. Compared to SAN, NAS provides lower performance. Furthermore, NAS provides only file sharing no other traditional file server
applications, such as email, database or printing servers, are supported.
The mentioned advantages and disadvantages suggest the area of
application to be midsized organizations with remote departments, or
branch offices of large organizations, with networks that have a mix of
clients, servers and operations. It is ideal for collaborative environments
with file sharing between departments. NAS solutions are successfully
implemented where there are requirements for easy file sharing, software development, computer-aided design and computer-aided manufacturing (CAD/CAM), audio-video streaming, publishing, broadcasting,
etc.
113
NAS implementation and converged infrastructure

There are several methods of NAS implementation. NAS can be implemented using a NAS appliance or filer, NAS blade or NAS gateway.
A NAS appliance combines a front end file server and back end storage system in a single unit. This approach is sometimes referred to as
a closed-box approach. In one device you get the file server and the
storage array. This solution is highly reliable and provides efficient performance. It enables easy installation, management and use. It is also
the least expensive NAS implementation. The downside is that NAS appliance is not scalable and does not provide pool storage, which makes
it difficult to achieve high utilization.
A NAS blade allows multi-protocol data storage in a large disk array. The NAS blade is basically an option that is installed to enterprise
level storage systems. It adds ports that make it possible to connect the
storage system to a local area network, and it has a built-in file server.
You can add more than one NAS blade option into your enterprise disk
array, which means that this solution is scalable. It also provides the full
functionalities of an enterprise storage system high performance, security, availability and redundancy (data can be replicated in addition to
RAID redundancy features).
In a NAS gateway, all devices communicate directly with the file system. A NAS gateway overcomes the limitations of a NAS appliance. It
separates the file server from the storage device, it offers better utilization rates and, most importantly, it allows creation of a converged NAS
and SAN infrastructure by providing NAS functionalities to SAN storage.
Combination of NAS with SAN capacity enables you to meet growing
storage requirements. A NAS gateway controller uses Fibre Channel
protocol to connect to external storage.
114
There are three types of storage networking architecture: DAS, NAS

and SAN. In DAS, a storage device is directly attached to an individual
server. SAN is a high-speed network of shared storage devices. SAN implementation uses Fibre Channel technology or iSCSI and operates on block
level. In NAS, data is presented at file level and the storage device directly
connects to the LAN.
See Figure 3.11 for comparison.
Users
LAN
NAS Gateway
Hosts
SAN
Raid Storage
Figure 3.10 Converged SAN and NAS infrastructure. NAS head with storage connected over
SAN. NAS scales to the limits of SAN, which is limited by the NAS file systems capacity. Implementation of several NAS gateways is possible; this type of solution is scalable. Converged solution can coexist with application servers, and it can be centrally managed.
DAS
NAS
Application
File System
Direct
Connected
SAN
Application
File System
IP Network
File System
SAN
Figure 3.11 Storage

networking architecture overview.
115
Network protocols
A protocol is a set of rules that governs communication between
computers on a network. It regulates network characteristics in terms
of access method, physical topologies allowed in the network, types of
cable that can be used in the network and speed of data transfer. The
different types of protocols that can be used in a network are:
Ethernet
Fibre Channel Protocol (FCP)
Fiber connectivity (FICON)
Internet Protocol (IP)
Internet small computer system interface (iSCSI)
Fibre Channel over IP (FCIP)
Internet Fibre Channel Protocol (iFCP)
Fibre Channel over Ethernet (FCoE)
Ethernet protocol uses an access method called carrier sense
multiple access/collision detection (CSM/CD). Before transmitting,
anode checks whether any other node is using the network. If clear,
the node begins to transmit. Ethernet allows data transmission over
twisted pair or fiber optic cables and is mainly used in LANs. There
are various versions of Ethernet with various speed specifications.
The Fibre Channel protocol defines a multi-layered architecture for
moving data. FCP packages SCSI commands into Fibre Channel frames
ready for transmission. FCP also allows data transmission over twisted
pair and over fibre optic cables. It is mainly used in large data centers
for applications requiring high availability, such as transaction processing and databases.
116
FICON is a protocol that connects a mainframe to its peripheral devices and disk array. FICON is based on FCP and has evolved from IBMs
implementation of Enterprise Systems Connection (ESCON).
Internet protocol is used to transfer data across a network. Each device on the network has a unique IP address that identifies it. IP works
in conjunction with the TCP, iSCSI and FCIP protocols. When you transfer
messages over a network by using IP, IP breaks the message into smaller
units called packets (third layer in OSI model). Each packet is treated as
an individual unit. IP delivers the packets to the destination. Transmission control protocol (TCP) is the protocol that combines the packets
into the correct order to reform the message that was sent from the
source.
iSCSI establishes and manages connection between IP-based storage
devices, and it hosts and enables deployment of IP-based storage area
networks. It facilitates data transfers over intranets, manages storage
over long distances and is cost-effective, robust and reliable. As we already mentioned, iSCSI is best-suited for web server, email and departmental business applications in small to medium sized businesses.
Fibre Channel over IP is a TCP/IP based tunneling protocol that connects geographically distributed Fibre Channel SANs. FCIP encapsulates
Fibre Channel frames into frames that comply with TCP/IP standards.
It can be useful when connecting two SAN networks over the Internet
tunnel, in a similar fashion to virtual private networks (VPNs) allowing
connection to a distant LAN over the Internet.
iFCP is again TCP/IP based. It is basically an adaptation of FCIP using routing instead of tunneling. It interconnects Fibre Channel storage
devices or SANs by using an IP infrastructure. iFCP moves Fibre Channel data over IP networks by using iSCSI protocols. Both FCIP and iFCP
provide means to extend Fibre Channel networks over distance. Both
these protocols are highly reliable and scalable. They are best suited
for connecting two data centers for centralized data management or
disaster recovery.
117
Fibre Channel over Ethernet is an encapsulation of Fibre Channel

frames over Ethernet networks. This allows Fibre Channel to use 10Gb
Ethernet networks while preserving the Fibre Channel protocol. FCoE
provides these advantages:
Network (IP) and storage (SAN) data traffic can be consolidated
using a single network switch.
It reduces the number of network interface cards required to
connect disparate storage and IP networks.
It reduces the number of cables and switches.
FCoE reduces power and cooling costs.
To put this into simple words, you can build your SAN using Ethernet
cables (mostly twisted pair). You can use one switch for your IP based
network traffic (LAN) and for creating SAN infrastructure. Even though
the switch and cabling are the same, LAN will run on TCP/IP while SAN
runs on FCP.
Storage area network security

SAN networks benefit from inherited security features over IP-based
networks. These features are determined by a smaller number of nodes
in the network and isolation from local area and wide area networks.
However, with the emergence of converged network solutions and
cloud computing solutions, it is necessary to employ some additional
techniques to ensure integrity and security of SAN infrastructure.
LUN mapping, which is sometimes also called LUN masking, provides
the means for permitting access to the particular logical unit from hosts
based on their physical address. Usually only one server can access one
LUN, either through one or two paths. In a server cluster (server redundancy, where a clustered server is ready to take over if the primary
118
server fails) it can be two servers accessing a single LUN. It is possible

to provide access to a LUN from more servers, but this is not usually
implemented in order to maintain consistency of data stored on a LUN.
In other words, when a server stores a database on a LUN, this database
is constantly changing because there are ongoing transactions. Furthermore, a LUN operates on the block level the file system is installed
on the server. These are the reasons data on this particular LUN with a
database cannot be read on another server from the perspective of
another server, the data is inconsistent and thus unusable. The LUN is
mapped on a disk array controller level using disk array management
software. The particular LUN can be accessed from servers, whose
HBAs WWN addresses are listed in disk array configuration software.
Access to the LUN is denied to all other WWNs.
LUN mapping is the feature fully managed on storage systems. Fibre Channel switches also provide a feature that helps to create a secure SAN environment. On the level of Fibre Channel switches, we employ zoning. Zoning is one of the most important mechanisms within a
switched fabric network that helps to achieve a higher level of security.
Zones are defined in order to establish rules that govern mutual communication of network devices. A Fibre Channel switch allows frame
forwarding only within the configured zone. Imagine that you have a
SAN infrastructure that consists of a disk array, tape library, five servers
and one backup server. Zoning can be designed in such a way that all
the servers are able to communicate with the disk array, while the tape
library is able to communicate only with the backup server. Zoning is
automatically propagated to all FC switches within the network, so it is
enough to configure just one switch.
Zoning can be configured as zoning based on WWN addresses, port
zoning or mixed zoning. Zoning based on WWN addresses allows communication between a particular disk array front end port and a server
host bus adaptor based on the WWN address. This type of zoning is
sometimes called soft zoning. It allows great flexibility because it is resistant to changes in cable infrastructure. This means that if you need to
add a new node and change cabling (e.g., which cable is connected to
119
what port in a switch) you do not have to reconfigure zoning. The WWN
address is a physical address, which means it is built-in and does not
change. On the other hand, in port based zoning, we do not use WWN
addresses; we define which ports of a switch allow mutual communication instead. This is sometimes called hard zoning. Compared to WWN
zoning, port zoning is more secure. But if you want to change cabling,
you have to reconfigure zoning.
Storage
System
FC Switch
2
Servers
Figure 3.12 An
example of WWN
based zoning.
Mixed zoning combines both port zoning and WWN based addressing. When we set up a mixed zone, we define which ports located on
the FC switch can communicate, and then, using WWN, we authorize
the particular host bus adaptor to access the zone. Mixed zoning is the
most secure. Implementation of certain types of zoning depends on security requirements and the number of installed nodes.
Additional tools that can help us make SANs more secure include the
definition of policies and standards and regular auditing. It is a matter
of course that all node microcode and firmware are updated regularly,
120
access to all node configuration tools is secured by using strong passwords and event logs are stored in a secure location.
Both LUN masking and fabric zoning should ensure that SAN resources
are accessed only on a need to access basis. This means that a server
can access only its own assigned LUN even though the switched fabric
allows practically all servers to all storage device connections. LUN mapping and fabric zoning cannot be configured from the production servers, eliminating possible intrusion threats. In LUN masking, a logical unit
is paired with the servers host bus adaptor WWN address. LUN masking
is configured on the storage system controller. In zoning, we can apply
soft zoning, hard zoning or a combination of both. Zoning is configured on
Fibre Channel switches and prevents frame forwarding to nodes that are
not members of the particular zone.
Network Attached Storage Security

Unlike SAN, which is to a certain degree isolated from LAN and WAN
environments, NAS is directly connected to the existing LAN infrastructure, which makes it more exposed to possible threats. Since NAS devices contain a file system and allow direct access to the files, NAS security is based on the definition of access policies and authentication.
Authentication services can be provided by NAS device management
software. However, in a business environment application, authentication is usually provided by the server with an installed directory service, such as openLDAP or Microsoft Active Directory. The server must
make the authentication service available on the network and it must
be powerful enough to handle authentication load. Users who are not
assigned necessary privileges are denied access to the particular folder
or file. Another way to enhance the security of a NAS device is to create
a virtual local area network (VLAN), which is a physical switched LAN
(or more LANs) divided into logical segments according to usage. We
can create a virtual network that will use only those local area network
121
resources that are needed by a certain group of users. For example, you
can create VLANs for company departments, project teams, subsidiaries, etc. To a certain extent, VLAN segmenting can resemble fabric zoning used in SAN because both are configured on the switch level. Access
to a particular VLAN is granted according to which switch port is used
or according to MAC address. Joining a VLAN can also be governed by a
special server that can grant access based on authentication.
This chapter provided you with a technological description of networking. Now, you should be familiar with storage system networking possibilities. Remember that modern storage systems are usually implemented in
a SAN environment, i.e., they are connected to Fibre Channel switches and
HBAs installed in servers. NAS devices connected directly to a LAN are also
increasingly popular. On the other hand, DAS and iSCSI connection implementation is increasingly marginal. The most important protocols are FC,
TCP/IP, iFCP and FCIP. FICON protocols are used only with mainframes.
122
Business continuity and

replication
How to describe the concepts of business continuity
How to describe business impact analysis and risk
assessment
What common backup strategies are and how they are
implemented
How business continuity and disaster recovery
are connected to IT
The meaning of replication and the kind of replication
possibilities we have
How to describe concepts of clusters and geoclusters
124
4 Business continuity and replication
Business continuity concepts

The major focus of this chapter will be on protection of our critical
data. We will discuss the techniques and tools we can use to ensure we
do not lose data that plays a crucial role in our business. However, even
before that, we will start with business continuity principles that create
a functional framework that encompasses not only securing data and IT
infrastructure, but continuity of the mission critical business processes
in non-standard and potentially destructive situations.
We all hope that our organization will not be affected by a cyber
attack or theft. However, these things happen and it is better to be prepared for them. Imagine there is a fire in your company that destroys
half the building your IT equipment and offices. To be prepared for
such a situation means that you have an alternative location, which can
be temporarily used as offices. It means that you need to know what
equipment is the most important and you have to be able to contact
your suppliers and order this new replacement equipment as soon as
possible. It also means that you must have employees trained and prepared for handling such situations. In the event that your servers and
storage systems are out of order, your applications do not work. These
applications may be responsible for accounting, ordering goods or shipping goods. If you are unable to process orders from your customers,
you could face significant financial loss. However, it is not only immediate financial loss; the whole incident, depending on its severity, could
seriously damage the reputation and image of your organization and
affect your future revenue.
Business continuity management identifies critical business processes and establishes rules and procedures that should prevent interruption of these key processes in the event of unexpected situations that
could negatively impact an organization. The goal is to reduce the negative impact of crisis situations on the organization and to mitigate the
risk involved in your area of business. The implementation of business
continuity management is governed by regulations and standards. In
125
most countries, it is a legal requirement to have a plan for handling crisis situations. The standards provide methodology that helps to create
a functional business continuity plan.
The best known standard is the British Standard for Business Continuity Management (BS25999), which provides terminology, as well as
guidance for the determination of business processes and their importance. Training based on this standard also provides you with best practices that can significantly simplify business continuity management
implementation in your organization. When you meet all the criteria
represented by BS25999, you can receive certification from an independent external auditing company, which proves that your organization is well prepared for handling the most probable incidents. This can
bring you an advantage over your competitors, not only through an improved response to times of distress, but also through increased credibility and trustworthiness. Business continuity management also helps
you protect your trademark, shows you how to make your investments
and can grant you lower insurance prices.
There are other well known and commonly used standards in addition to BS25999, such as the North American Business Continuity Standard (NFPA1600) and BS ISO/IEC 27001-2:2005 International Information
Security Standard. It is important to note that, according to these standards, business continuity principles are applicable to any organization,
regardless of its business area and size.
Business continuity planning

Business continuity (BC) planning is the implementation of business
continuity concepts with respect to the particular organization. Its obvious that the business continuity plan of an international airport will
be very different from the business continuity plan of a supermarket
since the processes of both companies are completely different. How-
126
ever, to prepare a functional BC plan, both organizations need to undergo certain procedures based mainly on analysis of their inner and outer
environments. The output of these procedures is then represented by
several documents that are all part of the BC plan. The functional business continuity plan includes:
A business impact analysis is the assessment of crucial business
processes of a particular organization.
A risk assessment is the determination of possible risks and the
attitude an organization assumes towards these risks.
The definition of policies and roles is based on the risk assessment output. A BC policy should define the focus of the BC plan
with respect to the organizations business strategy and the current market state. The definition of roles provides transparent
and clear division of responsibilities among BC team members.
A recovery resources description is also a vital part of the BC
plan. An organization can have its own resources in several locations or these resources can be provided by suppliers and contractors.
A business recovery plan is a step-by-step plan, or script, that defines procedures to be taken when a particular situation occurs.
A disaster recovery plan is a recovery plan for IT infrastructure
and equipment. It describes how to get the critical applications
working again after an incident takes place.
A testing and training schedule and maintenance plan provide
methodology and a timeline for business continuity plan testing.
It should also include the name of a person responsible for keeping the BC plan updated.
The first step is to set up a business continuity team, which usually
includes business processes owners and upper management. This team
must translate the business requirements into an overall business continuity plan that includes technology, people and business processes for
127
recovery. The major considerations are requirements for data restoration (such as whether the business requires restoration up to the point
of the disaster or can restore from a previous data backup) and recovery
time requirements. These considerations determine the technologies
and method used to support the disaster recovery plan. For example,
if the business requires near-continuous recovery of data (with no lost
transactions), it will likely use remote mirroring of data and wide-area clusters that enable a hot standby application environment. The
shorter the recovery time and the less transaction loss, the higher the
cost of the recovery solution. As more and more businesses rely on
critical applications to generate revenue, the requirement increases
for shorter recovery time (less than 1hr) or even continuous recovery
(within minutes).
We have described a business recovery plan, which includes scripts
that provide step-by-step instructions on how to handle particular situations. There can be anywhere from two or three scripts to dozens of
scripts prepared for different possible scenarios, depending on the size
of an organization. Scripts are prepared for handling events such as fire,
burglary or sabotage. When such an event occurs, the business continuity team activates the right script and simply follows the instructions
divided into tasks. A well prepared BC plan significantly increases an organizations resistance to unexpected situations. All BC documentation
needs to be regularly updated and audited to ensure that the business
continuity plan reflects the current state of inner and outer environments for the particular organization.
Business impact analysis

A business impact analysis (BIA) is part of the planning phase that
identifies which business processes, users and applications are critical
to the survival of the business. This process feeds into the next planning
step for business continuity. Performing a BIA can help identify cost
128
of data unavailability (downtime), which can be used to support decisions for various recovery solutions.
A BIA provides information necessary for creating BC strategies. Its
extent is determined by the scope of products and services the organization provides. Business processes can be divided into operation processes (such as production, sales, customer support, distribution and
billing), support processes (such as IT, human resources and external
services) and strategic processes (such as management, project management and planning). If the importance of a process changes over
time (e.g., accounting at the end of financial year), assessment is made
according to the busiest period. The mutual relationship and dependencies of the processes also need to be considered.
The output of a business impact analysis is expressed through the
following values: maximum tolerable period of disruption (MTDP), recovery time objective (RTO) and recovery point objective (RPO). These
values are attributed to each process and process support layers. These
values are very important and basically tell us what level of redundancy
and availability we need and how costly it will be to meet these requirements. RTO and RPO are major concerns when deciding what backup or
replication strategy we must use.
The recovery time objective value tells us what the tolerable duration of the process disruption is. In other words, how soon we have
to resume the process to avoid significant negative consequences.
RTO is usually determined in cooperation with the business process
owner. We can also set RTO for process dependencies; in this case we
are talking about RTO layers, which can be, for example, resumption
of power supply, air conditioning in the server room, resumption of
server operation, network or storage system, RTO of installed application.
The recovery point objective, on the other hand, tells what data loss
we can afford for the particular business process. In other words, it tells
us how old our backup data can be. If the last backup is at 7:00 PM and
129
our storage system is destroyed at 10:00 PM by fire, it means we lose all

the data that was stored in those 3hrs.
Maximum Tolerable Period

of Disruption (MTPD)
Financial loss
>500 t.
<500 t.
risk acceptance level
<250 t.
<100 t.
<50 t.
<10 t.
process 1
1/2
MTPD
30
Process disruption length (days)

Figure 4.1 Maximum tolerable period of a business process disruption.
The maximum tolerable period of disruption is output of the business impact analysis, but risk management also plays a role when determining MTDP values. The MTDP value shows us the length of a business
process disruption after which the viability of this particular process,
or organization as whole, is seriously threatened. In other words, once
the time marked by MTDP value passes, it is practically impossible to
restore the business process because the damage caused by the disruption is too severe. The MTDP is determined on the basis of financial
loss incured because the business process was interrupted, and the risk
acceptance level, which marks the amount of financial loss an organization is capable of handling. To avoid reaching the MTDP boundary, an
organization must take precautions. The best option would be to take
such precautions that the financial loss is near to negligible. However,
130
high availability and failure resistant systems are so expensive that it

is necessary to determine the risk acceptance level. Up to this level,
financial loss can be significant, but it does not endanger the business
as whole. Accepted possible financial loss is still significantly lower than
the cost of implementing precaution measures. Remember, that the
first letter of MTPD stands for maximum. This makes the MTPD values
highly hypothetical because no company will set its recovery time objectives as high as MTPD. Process disruption time allowed by recovery
time objective values will always be significantly lower than values determined by MTPD.
Infrastructure recovery
Data recovery
Applications recovery
Services recovery
Process recovery
Emergency
operation
Replacement procedure
RPO
Disaster
Regular
process
operation
RTO
MTPD
Regular
operation
Time
Figure 4.2 Business process recovery diagram.
In Figure 4.2, you can see that the RPO is always lower than the RTO.
We also know that the RTO is lower than the MTPD. The resumption of
a disrupted business process is achieved by completing individual recovery steps. First, it is important to recover the infrastructure to have
power and cooling for servers and disk arrays. Then we can recover the
data from backup. The technology nowadays allows us to set the RPO
value to zero and the RTO to seconds or minutes. This can be achieved
by redundant duplicate infrastructure (servers and disk arrays) placed
in a separate location, which is ready to take over when the first location fails. This type of solution is called geoclustering, and it is by far
131
the most expensive solution available. We will discuss geoclusters in

detail when we talk about cluster solutions and remote replication. If
the business process is not critical and its not necessary to invest large
amounts of money into geoclusters, we can set RPO and RTO values to
minutes and hours. This setting is allowed through several means, such
as synchronous remote data replication and an alternative location. The
cheapest solutions based on tape backup allow RPO and RTO to be set
to hours and days.
When we design an IT infrastructure, we need to consider RPO and
RTO values. RPO and RTO values are concerned with business processes
that are dependent on hardware availability, software functionality, human intervention and non-standard situations (sabotage, natural disaster, theft, etc.). When everything is working as it should work, we say
that the system is available, meaning users can access applications running on servers and servers can access data stored on disk arrays. Availability is expressed in a percentage that shows us how much time a year
the system is operable. Availability takes into account unexpected incidents causing disruptions, as well as planned maintenance procedures.
If we have 99.9% availability, it actually means the system will be out
of order for a maximum of 9hrs a year. Other commonly used figures
include 99.95% availability, which represents a total outage of 5hrs a
year; 99.99% availability, which makes 1hr a year; and 99.999% which
allows disruption for only 5min a year. The availability reaching higher
values than 99.9% is commonly called high availability.
Midrange and especially enterprise level storage systems are designed to provide high availability. High availability is achieved by redundant dual controller architecture and by hot-swappable components
that can be replaced without disruption. Storage systems usually also
allow nondisruptive maintenance, which includes microcode upgrade
and other procedures, such as migration and logical unit expansion.
132
Risk assessment
Risk assessment is another part that constitutes the proper business
continuity plan. It provides important data that amends business impact
analysis and helps to make the right choice of means when designing
failure resistant IT environment. Risk assessment determines the probability of threats that can lead to disruption of an organization, and it also
determines the impact of these threats. The difference between BIA and
RA is that BIA determines the impact of business process disruption on
the operation of an organization, while RA measures and evaluates the
possibility of threat occurrence and its impact on the business processes.
The major concerns are the probability of threat and the impact of
this threat. Threats can be represented by a human factor failure, a natural disaster, a technological failure, an economic situation or political
instability. Another categorization divides threats into random and intentional threats that are possible to prevent and threats that are not
possible to prevent. Some threats can be recognized before they strike,
which buys us some time to prepare for them, and other threats have
no prior notice. It is also necessary to determine whether the particular
threat can strike locally (e.g., fire) or regionally (e.g., earthquake). All
these factors need to be accounted for, and this information, in cooperation with a business impact analysis, helps us choose a functional
business continuity strategy.
The impact of a threat can lead to financial loss, unavailability of
employees, limited or no access to a building and premises, hardware
failure, data unavailability, infrastructure outages and unavailability of
critical documentation. If you took classes in risk management, you
probably know there are many methods that help us assess and evaluate the risks. Among the most important are the what-if analysis, the
cause-consequence analysis, the brainstorming approach and the event
tree analysis. For the purpose of this book, all you need to know is that
some of these methods are more sophisticated and scientific than others, and good risk management usually makes use of a combination of
different methodologies.
133
The purpose of risk assessment is to help an organization assume the

right attitude towards a risk based on its probability and the severity of
its impact. The possible threats and calculated probability constitute a
risk quartile matrix, which looks something like this:
High
Insurance
Outsourcing
Risk Elimination
Low
Risk
Acceptance
Risk Mitigation
Precautions
Low
High
Financial loss
Probability
Figure 4.3 Risk assessment matrix.
If the threat is unlikely to happen and would cause low financial loss,
an organization usually accepts the risk. The threats that are unlikely
to happen but would cause high financial loss (e.g., a fire or flood) are
usually treated with the proper insurance. If the threat is likely to occur and cause low financial loss, an organization will try to mitigate the
risk by taking precautions. When the threat is likely to take place and
the financial loss that would be incurred is high (e.g., your organization
is set up on a river bank that floods every spring), the risk needs to be
eliminated.
From the perspective of the IT infrastructure and the continuity of
its operation, we are mostly concerned with threats with impact that
can be avoided by taking precautions. The precautions to be taken are
summed up in a risk treatment plan, which is part of risk assessment
documentation. The risk treatment plan determines clearly what precautions and actions will be taken to prevent the threat or to mitigate
the threat impact.
134
RPO and RTO are major considerations when designing a failure resistant IT environment. Remember that RPO stands for recovery point objective, and it tells us what amount of data transactions we are allowed to
lose in the event of disaster. It determines how old our backup data can
be. RTO, which stands for recovery time objective, tells us what the maximum desirable period of disruption is. After the RTO time passes, the application should work properly. The MTDP, or maximum tolerable period
of disruption, is purely hypothetical and it defines the point of no return.
The RTO will always be significantly lower than the MTDP.
All these values are outputs of the business impact analysis. The risk
assessment complements and amends the BIA.
Business continuity vs. disaster recovery

The business continuity plan focuses on all business processes important for a particular organization. Imagine you need to prepare a BC
plan for an airport. This BC plan will include a description of actions that
need to be taken, for example, when its freezing outside and the supplier failed to deliver fluid for defrosting the aircraft. As you can see, this
has little to do with IT infrastructure and its operations, yet it definitely
needs to be accounted for. BC planning is a large-scale process that explores how a particular organization is operating. A BC plan can be tailored even for companies that do not use IT equipment at all, although
there would be very few of them nowadays.
Now its time to stress that disaster recovery is not business continuity. Sometimes these terms are mistakenly confused and their use
is not interchangeable. Disaster recovery is the part of business continuity that focuses only on IT infrastructure and business processes
that depend on IT equipment. Some organizations have all their business processes depend on IT infrastructure, for example, data centers,
mobile network operators and banks. For these organizations, disaster
recovery is crucial. However, even these organizations need a business
135
continuity plan that considers supplier and employee availability in

times of crisis.
Some storage system manufacturers see business continuity as the

equivalent of disaster recovery. You can see this in the name of their management software or in their marketing. We should now understand that
while this can be true from their perspective, in a broader context, business continuity will always stand above disaster recovery.
Disaster recovery plan

Business continuity should prepare your organization for a disaster
that can cause disruption of vitally important business processes. It allows you to learn a lot about your own company, and it allows you to
recognize possible threats before they even happen. A disaster recovery plan, together with a business recovery plan, will help you most
when a threat strikes. A business recovery plan generally focuses on the
business processes. It provides solutions in situations when the building with your offices is inaccessible, your employees cannot get to work
because of harsh weather conditions, and other disruptive events. A
disaster recovery plan focuses only on the continuity of business processes that are dependent on IT infrastructure. By IT infrastructure, we
mean servers, switches, disk arrays, connectivity to the WAN (Internet),
etc. We can also include power supply and air conditioning. In a situation of distress, when your IT equipment is stolen, destroyed by fire or
malfunctioning, you will use your DR plan, which contains a description
of all IT related processes together with step-by-step guidance for their
recovery.
136
A disaster recovery plan contains:

Basic information: purpose, area of application, requirements,
log of DR plan modifications, members of DR team and their roles
and responsibilities
Notification/activation phase: notification procedure, call tree,
damage assessment, activation criteria, plan activation
Recovery procedures: succession of recovery procedures according to their importance, logging, escalation
Standard operation resumption: checking whether all systems
work properly, termination of DR plan
Amendments: call book, vendor SLA, RTO of processes
Lets start by talking about the notification/activation phase, which is
especially interesting. When something happens, lets say, part of your
equipment gets stolen, you need to notify the people responsible for
recovery of processes dependent on IT equipment and IT infrastructure. You would use a call tree list, which tells you who should be notified first. Together with the members of the DR team, you would assess
the extent of damage and the expected recovery time. This assessment
must be done quickly because the clock is already running.
To help you with damage assessment, the DR plan provides individual IT equipment locations in the building and a checklist that guides you
through the procedure. The checklist is usually designed so that even
employees with no IT experience can assess the damage. The checklist typically uses simple directions, such as: Check switch x located in
room y. Are the green lights glowing? The instructions and questions
are this simple so that unqualified personnel can perform the assessment before qualified personnel arrive. Once you have analyzed the
extent of the damage, you can check whether the damage is serious
enough to qualify for DR plan activation. If yes, the plan can be activated. Once the DR plan is activated, all members of the DR team, or
their delegates, must be present and ready to perform their part of the
recovery procedures.
137
Recovery procedures are followed in a logical order, where the most

important processes need to be recovered first. The importance of a
procedure is usually assessed based on prerequisites if you have not
restored Internet connectivity, then it is useless that your email server
works seamlessly. You should also consider which procedures can be
done simultaneously.
The order of recovery procedures in a small company can look similar to this example:
LAN recovery
WAN recovery
Domain controller
Email server
SAN recovery
Backup server
Microsoft SharePoint server
Accounting software and databases, etc.
Each procedure has its own recovery time estimate, checklist, and
supplier and maintenance provider contact details (escalation). Once
all the procedures are complete, the functionality of all systems can be
tested. When everything is operational, the DR plan is terminated.
Speed of individual process recovery is determined especially by
technological means. The performance of disaster recovery planning
depends highly on backup strategies, data replication strategies and
cluster solutions. Some solutions are less resistant to failure than others, but they are usually less expensive.
Disaster recovery support in storage systems

Disk arrays store our data, which makes them the most vulnerable
part of our IT infrastructure. In the event of a disaster, you can always
buy new hardware, and you can install your application from a DVD
or some other media, but you cannot recover your data if you did not
have it protected effectively enough. Thats why storage system manu-
138
facturers offer a wide range of services and software solutions and

best practice guidance to make storage systems even more resistant
to data loss.
We can say that storage systems are inherently redundant. This
means that they provide no single point of failure architecture, where
all important hardware is installed twice (remember, CTL0 and CTL1).
No vendor would sell storage without these capabilities. Nor would a
vendor sell a storage system without RAID support. This is what makes
storage systems reliable and powerful. However, software also plays an
important role. Software can help you employ and optimize data replication. Software agents can also enable close cooperation between
applications running on production servers and disk arrays. The supplied software will also help you to maintain data consistency and integrity, which is especially important for DR procedures. If your data is
not consistent, it cannot be read and is, therefore, useless. It is essential
to maintain consistency, especially with databases. Databases are living
data, changing every second due to a high number of ongoing transactions. Software supplied by a storage system manufacturer can help you
simplify management of very complex operations. Remember that if
you purchase a storage system, the actual hardware may cost less than
the software licenses. This should give you some idea of how important
software supplied by a storage system manufacturer is.
Backup and replication

By now you probably understand that both data backup and data
replication are key words in disaster recovery planning. Before we get
to the description of the individual techniques and solutions, its necessary to understand what we mean when we talk about backup and
data replication. In other words, well discuss what the difference is
between these two.
139
In broad terms, a backup means we create duplicates of data, either

in the form of files or raw data blocks. Data replication can be seen as
part of data backup. From home computer use, you know its good to
keep a copy of your important files in another location, such as another
hard drive, a USB flash drive or a CD. If your computer breaks and you
lose your data, you can restore it from your backups. This is the basic
idea. In an enterprise environment, there is so much data that its impossible to manually back up all the files. With a server application,
its even more complicated because you have to back up very complex
databases. That is why there are many dedicated hardware devices and
software products to handle the process in an effective and reliable
way.
In technical terms, data replication is very different from data backup. When we back up data, it means that if we need to access this data,
we have to restore it. The backup restoration, therefore, takes time.
Depending on the media that contains our backed up data and depending on our backup strategy, the process of data recovery can take from
hours to days. On the other hand, data replication does not need any
recovery; replicated data can be used instantly, or nearly instantly.
When you back up data on an enterprise level, you use dedicated
software and equipment. When we want to restore this data, we must
use this software and equipment to make the data accessible to an application that is running on your production servers. The data replica
can be accessed directly by applications running on your servers; there
is no need to use backup servers and software.
You may wonder why we back up our data, when data replication
is clearly better. It is all about cost and capacity. You can back up your
data to a tape library, which is quite a cheap solution and very reliable. Tapes can hold terabytes of data. For data replication you need
to have a storage system and enough disk capacity installed in this
storage system. Imagine that you create a LUN on a disk array. You
want to replicate the data on this LUN. To do that, you need to create
another LUN with exactly the same size as the production LUN. Then
you create a replication pair out of these two LUNs. The source LUN
140
becomes a P-VOL (stands for production, or primary, volume) and the

LUN that contains the replica becomes an S-VOL (secondary volume).
In the context of storage systems, data replication is controlled solely
by a disk array. An S-VOL contains exactly the same raw data blocks as a
P-VOL. The consequence of this can be understood in another example:
imagine you create a LUN of 10GB. This LUN may contain only 1GB of
data. You will create a replication pair, the source LUN will become a PVOL and the target LUN will become an S-VOL. Both volumes will have
the size of 10GB and both volumes will contain only 1GB of actual data.
This means that data replication also copies the empty blocks that contain no real data. This is what makes replication more expensive than
data backup. The major downside of data replication is that it merely
mirrors the data on the P-VOL. If the P-VOL becomes corrupted or part
of the data is accidentally deleted, data replication will mirror this corrupted data.
Both data backup and data replication play important roles in data
loss prevention. Whether we employ backup or replication depends on
the data lifecycle and the importance of a business process that uses
the data. Less critical data can be backed up because if there is trouble,
we do not need to restore it in 10min. Important data will be replicated
to make it accessible under any circumstances. Note that when we employ data replication, we usually back up this data as well. If a server
fails, replicated data is used. If a disk array containing both a P-VOL and
an S-VOL fails, we need to restore data from a back up. In an enterprise
environment, tape libraries are standard for data backup.
Backup concepts
We have been talking about backup for a while and now it is time
to learn how it is actually done. In your organization, you have several
141
production servers that run important applications. To back up the data

from these servers, you need:
A backup device: In an enterprise environment it will most likely
be a tape library, but it can also be a storage system with a LUN
dedicated for backup. (Do not confuse this with a replication
pair!)
A backup server: In most cases, you need a backup server that
is communicating directly with the backup device and controls
backup from all the servers.
Backup software: This software runs on a backup server and allows you to make configurations according to your needs.
Backup agents: These agents are small applications installed on
all your production servers. They are part of backup software and
they allow communication between a backup server and production servers with applications and databases.
The backup server can be part of your LAN or SAN environment, depending on your requirements. LAN based backup topology does allow
you to use either a tape library or NAS device as a backup location.
The advantages are easy implementation and low cost. The disadvantage is that backup creates extra traffic, which can slow down your LAN
performance. SAN based backup topology uses a backup server that is
installed in SAN environment, which enables it to take advantage of its
features fast Fibre Channel data transfer and inherited security.
A backup server then pulls the data scheduled for backup from production servers. It is also possible to configure backup agents to push
the data from a production server to a backup server. However, this is
not recommended because it is less secure. If an agent pushes data to
the backup server that means it has permission to do write operations.
The pull strategy is more secure because it eliminates the risk that the
backed up data will be overwritten, either accidentally or on purpose.
142
File server
Backup server
Microsoft Exchange server
Disk backup
Tape backup
SQL server
Figure 4.4 Backup over LAN, the simplest solution.

The backup server pulls data from production servers
and then sends data to a tape library or NAS device.
To choose our backup topology and the proper software, we have to

decide when and how we want to back up our data. We have to prepare
a backup schedule and define the time of a backup window. A backup
window is a time of day when we make our regular scheduled backup.
It is usually set at night when the systems are not under high load and
backup traffic will not result in decreased performance of production
servers and network.
In addition, we also have to choose to what extent the data is going
to be backed up. We can decide to create a full backup, which means
that every time a backup is scheduled, all data is stored in exact copies. This can be slow and demanding in terms of capacity needed for
backup.
Sometimes a full backup is understood as system imaging, meaning that you create an image of the hard drive installed in the server,
143
which contains production data. This image can then be used for recovering the server to the exact state it was in at the time of backup. This
technique is often used for recording known good configurations after
a clean install and for distribution of this configuration to other servers. A system image can be useful for backing up employee desktops
and notebooks, but it is not likely to be used as a tool for making ongoing backups of all the server applications and their databases. Other
backup models are by:
Incremental backup at the beginning, a full backup is made.
Then at every scheduled backup, only the data that was changed
or added since last time backup is recorded. The disadvantage is
that when you need to recover data, you need to combine the
initial full backup with all incremental backups which is demanding in terms of time and capacity.
Differential backup at the beginning, a full backup is made. At
every scheduled backup, only the data that differs from the initial full backup is recorded. This technique delivers good recovery
times you need to combine only the initial full backup with the
most recent differential backup.
Reverse delta backup at the beginning, a full backup is made.
At every scheduled backup, the initial full backup image is synchronized so that it mirrors the current state of data on servers.
During the reverse delta backup, a differential file is created that
allows you to go back and restore previous versions of data if
they are needed. This is probably the best option to choose, because it allows you to achieve the fastest recovery times since
you always have the most recent full backup ready to use.
The last thing that remains to be considered is backup validation. No
matter what backup model you choose, it will always be a structured
backup, meaning that backed up data contains additional information,
such as when the data was recorded, where, and what the extent of the
backup was. Backup validation is a functionality presented by backup
management software. It checks whether all data was backed up suc-
144
cessfully with no error occurrence. It is a self-check mechanism that

provides feedback to IT administrators. Backup management software
also offers plenty of other features, such as backup data encryption that
enhances the security of your data.
We didnt mention one backup concept closely connected with storage systems serverless data backup. This technique is sometimes called
extended copy and is based on SCSI command that allows you to backup
data from LUNs directly onto a tape library, without the participation of
any backup server. A serverless backup environment is therefore SAN
based.
Backup based on tape libraries

Magnetic tape backup is the cheap and safe way of depositing your
data for recovery in case of primary data loss. The tape library used in
midsized organizations usually costs a bit less than a midrange storage
system. The individual tapes offer capacity in terabytes and they are
much cheaper than hard drives. The important limitation is that data
stored on tapes is accessed sequentially. It is similar to old VHS video
tapes you need to rewind or fast forward the tape to get to the place
where your data is. Despite this drawback, magnetic tapes are still used,
not only because of their price, but also for their reliability. Magnetic
tapes can store data for many years. Another specific feature of tapes
is WORM functionality. WORM stands for write once read many, and
this functionality should ensure that backed up data cannot be altered
or modified once it is written.
When we back up on tapes, we need to determine how many copies
we need to meet our business continuity policy. Usually we create one
onsite copy and one offsite copy. The advantage of tapes is that they
are easily moved to another location. An offsite backup repository can
145
be located at another building owned by a particular organization or

at specialized data center that offers disaster recovery services. In the
event of a disaster, the first thing to do is restore servers you need to
either install OS or restore it from an image. After that, you can restore
your data from onsite or offsite tapes. Data recovery from tapes can
take hours or even days.
Backup strategies
When deciding what backup strategy to employ, we need to sum up
all our requirements and make sure these requirements are met. We
have to be sure we are protected against:
Disk failure if we use a storage system, then RAID redundancy
should provide a good level of security. In addition, we can employ a storage-to-tape backup strategy.
Accidental file deletion backup software running on a backup
server should take care of this problem. It is also possible to restore files from tapes.
Complete machine destruction we need a backup machine,
system image that will restore known and used configuration,
storage system, tapes, etc.
Destruction of any onsite backups we would need remote
replication and offsite tapes.
We also need to decide to what extent we will back up. A business
impact analysis lets us know the importance of business processes. This,
together with RPO and RTO values, is the key to determining our backup
strategy. You need to work with many variables how much storage
capacity is possible to dedicate to backups, time and money available
for backup software and technology, human resources allocation on
backups, and performance impact on production servers and network
146
bandwidth. The backup strategy needs to be designed as scalable to

continue meeting the backup requirements as the business grows. The
most important thing is to know what dangers you want to protect
against and how you will handle restoration procedures for each item
you backup, such as files, databases and system configurations. It is also
likely that you will employ different backup strategies for production
servers and workstations. Regular backup testing and validating helps
to determine whether the currently used backup strategy is working
sufficiently or whether it would be advantageous to make adjustments.
Another thing to consider is the backup data retention plan. You
need to define how long you will keep individual versions of backup
data. If your backup server creates backups every 2hrs, you need to set
the number of copies to be retained. When the limit is reached, the
oldest backup is automatically deleted. Data retention has a lot to do
with both legal requirements (e.g., those set by accounting, hospitals
and government institutions) and with the data lifecycle. As we know
from the first chapter, the data lifecycle represents the changing value
of data over time. It is necessary to make such backup settings that
take data lifecycle into consideration, because, otherwise, you may not
realize the data you are backing up intensively every two hours is not
so important anymore and you are wasting money and resources. Your
backup strategy, therefore, needs to be updated regularly.
Backup optimization
Every organization has limited resources, especially in terms of capacity that can be used for backup and the cost of backup hardware and
software. It is therefore important to use resources carefully to maximize the utilization of your equipment. There are several techniques
that can make backup more effective. The desirable backup solution is
fast, does not have much influence on production server performance
and network bandwidth, and offers good value for money. The tech-
147
niques that allow you to achieve better utilization are data compression, deduplication, multiplexing and staging.
Compression is well known from personal computers. Most users
have already encountered .zip or .rar files that represent a package of
files encapsulated into one archive to provide easier manipulation and
compression the output archive file is smaller than the total of the
original files. The efficiency of compression is expressed as a ratio. For
example if you have a compression ratio of 1:2, the compressed file
will be half the size of the original file. The compression is achieved by
coding the files with a special algorithm. The efficiency of the algorithm
is based on the level of information entropy presented in the source
file. Simply said, information entropy is represented by sequence of
numbers in code and their randomness. The higher the level of randomness, the higher the entropy. If we had code represented by totally random numbers with no hidden pattern, then compression is not
possible. This, however, does not happen very often, and usually it is
possible to achieve at least some compression ratio. Lossless data compression can be compressed to an extent allowed by the data entropy.
A lower amount of randomness allows the algorithm to find underlying
patterns that are logged. If the pattern occurs more than twice, the
second occurrence is replaced by a reference to the first occurrence.
Compression is employed in tape libraries without exception. Processing compression algorithms may affect backup performance. In tape
libraries, this problem is eliminated by adding extra hardware processors that take care only of calculating data compression and decompression. Compression therefore does not affect read and write speed
in tape drives.
Deduplication is the technique that eliminates duplicities in data.
Thanks to this technique, it is possible to save disk space. Unlike compression, deduplication searches for a larger string of identical data.
If we backup our email database, it is likely there will be dozens of
instances of the same email. If the duplicate data is found, it is replaced by reference to the original data. Deduplicated backup data
can occupy up to 90% less disk space than data that contains duplici-
148
ties. We can employ inline deduplication, which takes place in real

time when we back up data. If the backup device recognizes a chunk
of data that has already been stored, it makes reference to this previously stored data without storing the new chunk of data. As opposed
to inline deduplication, we can employ post-process deduplication,
which takes place after the backup is finished. First, backup data is
stored and then it is analyzed for duplicities. The post-process deduplication is more effective than inline deduplication, but it requires
more disk space. The extra disk space is required because, during the
analysis of backup data, deduplication software creates large temporary files. Deduplication is very popular and often implemented.
Note that deduplication cannot be used with tape libraries. The data
stored on tapes is accessed sequentially, which disables the possibility of easy manipulation with previously stored data (rewinding and
fast forwarding). Data deduplication can be used when we back up
our data on hard drives or storage system LUNs.
Multiplexing in the context of data backups expresses the ability of
software and equipment to back up data from several sources simultaneously. In broader terms, multiplexing is a method that combines
multiple digital data streams into one signal that can be transferred
over one medium (wire). Similarly, in relevance to backup optimization,
multiplexing allows you to backup all your IT infrastructure on a single
backup device, providing it has enough capacity and it supports this
functionality.
Disk staging is a key part of the disk-to-disk-to-tape backup procedure (D2D2T). When we want to back up our data to a tape library, we
have to consider whether the data processing speed of the particular
tape device is sufficient. To speed up the backup process, we can set the
software to back up to a hard drive (or LUN) first and then transfer the
data from this hard drive to tape. The hard drive therefore works as a
temporary storage area on a similar principle to cache. Disk staging can
increase performance of frequent and small random access restores,
because the disk has much faster random access than tape (no rewinding and seeking on the tape). If we need to restore data, it can still be
149
present on the staging disk, ready for fast restore (depending on the
backup software configuration).
All these techniques can be employed together, and they often are.
Remember that you cannot employ deduplication when using a tape
library as a primary backup device. Deduplication is, however, possible
when we also employ disk staging. Data is deduplicated on the hard
drive and then transferred to a tape library. This, however, adds an extra step when restoring the data data is loaded from the tape and
then it must be restored one more time by reversing the deduplication
process. Multiplexing and compression are common features offered by
most backup software manufacturers.
Replication objectives
In this chapter, we are describing techniques that ensure high availability of data and business continuity with a focus on disaster recovery
and data loss prevention. There is a large number of solutions that help
to accomplish these objectives. Their efficiency goes hand-in-hand with
their cost. The backup solutions we have discussed so far are among the
cheaper solutions. However, even backup solutions are scalable and can
deliver high performance. As you know, all backup techniques require
some time for restoration, which will always be their limitations. For
critical applications that need to be online all the time, backup itself is
not sufficient, because recovery time from backup does not meet RPO
and RTO requirements. For this case, we have data replication. On the
following pages, we will not be dealing with database replication or replication done on the application level. We will focus mainly on replication possibilities offered by storage system manufacturers. Remember
that all the replication techniques we describe in detail happen on the
controller level, without participation of production servers. We are
talking about array-based replication.
150
The reason we use data replication is to achieve lower RPO and RTO
of key business processes. When the data replication is employed together with server redundancy (server clusters) we can reach zero RPO
and RTO in minutes. When we add a hot standby site in another location, we can reach zero RPO and RTO in a few seconds. The data replica
contains the same data blocks as a source volume, which enables immediate usage of data by application with no restore procedures. Data
replicas are also useful for application testing and data migration. We
are going to start with in-system replication and and snapshots, which
can also be considered as a replication technique. We will see how
replication works with server clusters. Towards the end of this chapter
we will get to solutions that can offer 99.999% availability. These solutions are based on remote replication and geocluster technology. As
the chapter progresses, the solutions presented are going to get more
technical, sophisticated and costly.
The replication strategy depends on the importance of data and its
availability. Bank institutions that offer Internet banking services or
mobile service providers need the highest availability possible because
banking transactions and call logging are ongoing processes that will
always have zero RPO and RTO in seconds. Imagine a bank that loses
data on user accounts or a mobile service provider that cannot store
information about customer calls. Even a minute of disruption can pose
an unacceptable threat and can lead to devastating consequences. Data
backup is not powerful enough to prevent data loss on such a level. That
is why we employ more sophisticated and costly solutions data replication and geographically diversified IT infrastructure.
151
Business Application
Exchange/SQL/Oracle/DB2
Primary
volume
Secondary
volume
Figure 4.5 The basic depiction of data replication as it is supported in storage systems. The
Primary Volume (P-VOL) is a LUN mapped to a server. Data from the P-VOL is replicated to an SVOL that can be located in the same storage system as the P-VOL or in a storage system located
elsewhere. The P-VOL and S-VOL must be the same size.
Overview of replication commands

When we decide to employ data replication on the storage system
level, we usually use software supplied by the storage system manufacturer. Individual replication functionalities are usually subject to
licensing, so you usually have to pay to enable a particular replication feature. Replication management software typically offers both a
graphical user interface (GUI) and command line interface (CLI). A command line interface is very useful for replication procedure automation
because it allows you to write your own scripts. Commands that are
used to control data replication differ from manufacturer to manufacturer. To learn how to create a replication pair, you usually need to go
through storage system documentation or attend a specialized training. However, there are few commands that are common with most
152
storage system manufacturers. Among these commands, or replication

functionalities, are:
Pair create and initial copy
Pair split
Pair resynchronization
Reverse synchronization/restore
When we create a replication pair, the source LUN becomes a PVOL and the target LUN becomes an S-VOL. Immediately after the pair
is created, initial copy takes place. Initial copy creates an exact replica
of all data blocks on the P-VOL, including empty blocks. During initial
copy, the P-VOL is accessible for I/O operations, but the performance
can be affected. Some storage systems allow you to select the priority
of the initial copy process when creating replication pairs. The slower
pace minimizes the impact of the initial copy operations on system
I/O, while the faster pace completes the initial copy operations as
quickly as possible. The best timing is based on the amount of write
activity on the P-VOL and the amount of time elapsed between update copies. Under all circumstances, it is better to create replication
pairs when the system is not under high load.
Write operations performed on the P-VOL during the initial copy operations will always be duplicated to the S-VOL after the initial copy is
complete. The status of the pair is COPY while the initial copy operation
is in progress. The pair status changes to PAIR when the initial copy is
complete. Once the initial copy is finished, differential data is synchronized to the S-VOL. See Figure 4.6 for reference.
Note that its impossible to perform read or write operations on the
S-VOL while the replication status is PAIR. To make use of the S-VOL,
you need to stop the continuous volume synchronization by splitting
the pair.
153
Host
P-VOL available to Host
fot R/W I/O operations
INITIAL COPY
P-VOL
All data
S-VOL
Data bitmap
P-VOL Differential
Updates S-VOL after initial copy
Host
P-VOL available to Host
fot R/W I/O operations
UPDATE COPY
P-VOL
Different data
Write I/O to P-VOL during initial

copy - duplicated at S-VOL by
update copy after initial copy
S-VOL
Figure 4.6 When we create a replication pair, initial copy takes place. Once the initial copy has
been completed, the content of the S-VOL is synchronized with the P-VOL regularly. When the
S-VOL status is PAIR, its not possible to read or write from the S-VOL.
Pair split operation performs all pending S-VOL updates (those issued prior to the split command and recorded in the P-VOL differential bitmap) to make the S-VOL identical to the state of the P-VOL
when the suspend command was issued and then provides full read/
write access to the split S-VOL. While the pair is split, the system establishes a bitmap for the split P-VOL and S-VOL and records all updates to
both volumes. The P-VOL remains fully accessible during the pair split
operation.
Once the P-VOL and S-VOL replication pair is split, you can access the
data on the S-VOL for testing, data mining, performance test or backup.
When you want to resynchronize the suspended pairs, you start the pair
synchronization operation. During this operation, the S-VOL becomes
unavailable for I/O operations. The regular pair synchronization procedure resynchronizes the S-VOL with the P-VOL. It is also possible to use
reverse synchronization with the copy direction from the S-VOL to the
P-VOL. When a pair synchronization is performed on a suspended (split)
154
pair, the storage system merges the S-VOL differential bitmap into the
P-VOL differential bitmap and then copies all flagged data from the PVOL to the S-VOL. When reverse pair synchronization takes place, the
process goes the other way. This ensures that the P-VOL and S-VOL are
properly resynchronized in the desired direction.
Another common operation that can be performed on a replication
pair is pair status transition. If necessary, it is possible to swap P-VOL
and S-VOL labels. S-VOL then becomes P-VOL and P-VOL becomes SVOL.
When you want to use the data replication feature, you need to create
a target LUN that has the exactly same size (measured in blocks) as the
source LUN. You should create the target LUN in a RAID group different
from the one that contains the source LUN to avoid performance bottlenecks. You can also create the target LUN on slower hard drives, depending on your requirements. Then you can start pair create operation. The
result of this operation is a replication pair. The replication pair consists of
a P-VOL (source volume) and an S-VOL (target volume). When the replication pair is active, data on S-VOL is updated from the P-VOL. It is not possible to perform read or write operations on a paired S-VOL. To be able to
use the S-VOL, you need to split the pair. To resynchronize the P-VOL and
S-VOL, issue a pair synchronization command. The synchronization can
work in both directions according to your needs.
In-system replication: logical unit cloning

In-system replication allows creating a single, local replica of any active application volume while benefiting from full RAID protection. The
result of this procedure is a volume clone that can be used by another
application or system for a variety of purposes. Depending on the type
of storage system, you can create several S-VOLs per PVOL within the
same system to maintain redundancy of the primary volume. In-system
replication allows you to split and combine duplex volumes and pro-
155
vides you the contents of static LUNs without stopping the access. It
is nondisruptive and allows the primary volume of each volume pair
to remain online for all hosts for both read and write I/O operations.
When the replication pair is established and the initial copy procedure
is completed, replication operations continue unattended to provide an
internal data backup.
In-system replication automatically creates a differential management logical unit (DM-LU), which is an exclusive volume used for storing
replication management information when the storage system is powered down. For redundancy purposes, it is usually possible to configure
two DM-LUs. The data stored on the differential management logical
unit is used only during shutdown and startup of the system. Note that
the DM-LU does not store the actual changed data; it just has the metadata for the differential data.
When we split the replication pair, we stop the ongoing synchronization and we make the S-VOL accessible for other applications. The SVOL created through in-system replication can then be used for backup
purposes. This enables IT administrators to:
Execute logical backups faster and with less effort than previously
possible
Easily configure backups to execute across a SAN
Manage backups from a central location
Increase the speed of applications
Ensure data availability
To use in-system replication to produce full logical unit clones for
backup purposes makes sense because it works on the volume level.
There is no need for a backup server, and the traffic does not lower network performance. The storage system management software controls
the whole procedure. Data from the S-VOL can be used for serverless
backup to tape library. Aside from high-performance serverless backup,
156
the replicated volume can be used for data mining, data warehousing
and full volume batch cycle testing.
Data mining is a thorough analysis of structured data stored in databases. It uncovers and extracts meaningful and potentially useful
information. The data mining methodology is based on statistics functions, interpretation and generalization of mutual relationships among
the data stored in a database. Data mining can provide highly valuable
information for marketing purposes. Data mining is conducted by specialized software that is programmed with some degree of artificial intelligence to be able to deliver relevant results.
A data warehouse is a specialized database used for data reporting
and analysis. The data stored in a warehouse is uploaded from production servers and their databases. This could affect the performance of
the server and the mapped LUN. This effect can be avoided by implementing in-system replication and using an S-VOL for feeding the data
warehouse database. The data warehouse database provides powerful
search capabilities conducted by queries.
Testing and benchmarks can be conducted using an S-VOL to determine database performance and suggest changes in configuration. It is
risky to test new procedures and configuration on a P-VOL because it
can potentially lead to data loss and it can affect performance. An S-VOL
provides the ideal testing environment.
The S-VOL replica also provides hot backup for instantaneous data
recovery if the P-VOL fails. If this happens, the P-VOL and S-VOL statuses
are swapped, which makes the S-VOL immediately available to the production server.
157
Server A
Server B
Backup
server
Onsite
tape
SAN
Disk Array
Disk
Production data B
Production data A
Clone A2
Offsite
tape
Clone A2
Clone B1
Clone B2
Figure 4.7 In-system data replication in combination with backup infrastructure. In-system
replication mirrors the source volumes and creates its exact replica or clone.
Copy-on-write snapshots
Copy-on-write snapshot software is a storage-based functionality
for point-in-time, read-only backups. Point-in-time backups capture the
state of production volume at a particular time. With in-system replication, the whole LUN is mirrored or cloned. On the other hand, snapshots
store only differential data into a data pool. The data pool consists of
allocated physical storage and is used only for storing differential data.
One data pool can store differential data from several P-VOLs. A snapshot pair consists of a P-VOL and a virtual volume (V-VOL). The V-VOL
does not physically exist but represents a set of pointers that refer to
the datas physical location, partially in the pool and partially in the PVOL. Since only part of the data belonging to the V-VOL is located in the
pool (and the other part is still on P-VOL), copy-on-write snapshot software does not require twice the disk space to establish a pair in the way
in-system replication creates full volume clones. However, a host will
158
recognize the P-VOL and the V-VOL as a pair of volumes with identical
capacity. The copy-on-write snapshot internally retains a logical duplication of the primary volume data at the time of command instruction.
This software is used for restoration; it allows data to be restored data
to the time of snapshot instruction. This means that if a logical error occurs in the P-VOL, the snapshot can be used to return the P-VOL to the
state it was in at time the snapshot was taken. The duplicated volume
of the copy-on-write snapshot function consists of physical data stored
in the P-VOL and differential data stored in the data pool. Although the
capacity of the used data pool is smaller than that of the primary volume, a duplicated volume can be created logically when the instruction
is given. The data pool can share two or more primary volumes and
the differential data of two or more duplicated volumes. Copy-on-write
means that whenever the data in the P-VOL is modified, the modified
blocks are transferred to the pool to maintain the V-VOL relevance to
the point-in-ime the snapshot was created.
Read
Write
Read
Differential
Data Save
Virtual
Volume
P-VOL
POOL
Write
10:00
11:00
12:00
Figure 4.8 Copy-on-write snapshot. Notice that both the P-VOL and V-VOL are accessible for
I/O operations. Snapshots can be created instantly.
159
When a production server needs to write to its mapped LUN that is

part of a snapshot pair, the request to perform a write operation on the
P-VOL is sent to the storage system. Before the write is executed, the
block is copied to the pool area. The set of pointers that represents the
V-VOL is updated, and all the requests going through the V-VOL for the
original block are physically taken from the pool. From the perspective
of the host, the V-VOL (snapshot image) has not changed.
Snapshots are usually taken regularly, and they are part of a backup
strategy. The snapshots can be taken every hour for LUNs that contain
very important data, or every day. There is no need to schedule backup
windows, since snapshot creation does not affect network traffic or
storage system performance.
We use copy-on-write snapshots because they require less capacity
to maintain virtual copies. In-system replication that produces volume
clones for backup solutions raises the entire system price, as the storage requirement grows based on the number of copies required.
Snapshot created
on Wednesday
P-VOL
V03
Wednesday
V02
Tuesday
V01
Monday
Link
Data for Monday

and Tuesday is already
saved on Pool.
Link
Pool
Physical VOL
Virtual VOL (V-VOL)
Figure 4.9 Snapshots are taken regularly. In this case the snapshot was created on Wednesday. The new snapshot image will refer to the data on the P-VOL. As the data is modified on the
P-VOL, the set of pointers in V-VOL 03 will change to point to the data stored in the pool.
160
Copy-on-write snapshot
In-system replication
Provides a very quick

copy since it is a virtual
volume
Offers superior protection

because it is a complete
copy; immediate restore if
P-VOL becomes corrupted
Instantaneous
From minutes to hours,

depending on the size of
the P-VOL
Disk space used
Size will vary depending

on rate of data change
but will be far less than
the P-VOL
Exactly the same amount

as P-VOL (measured in
data blocks)
Data recovery
time after P-VOL is
corrupted
Two stage restore:

1. V-VOL verify, data not
available to host
2. Copy back from a
tape library
Instantaneous
Size of physical
volume
P-VOL is larger than pool

for one V-VOL
P-VOL = S-VOL
Features
Time to create
Pair configuration
Generally speaking, storage systems support a higher

number of snapshots than data clones
P-VOL can be restored
from any V-VOL
Restore
P-VOL can be restored

from S-VOL
Figure 4.10 Copy-on-write snapshot and in-system replication comparison
Server clusters
To create completely failure resistant IT environment with zero RPO
and nearly zero RTO, it is not enough to employ data replication and
backup. Data backup requires time for data restore. Data replication
provides instantly available data if the P-VOL becomes corrupted and
fails. Neither of these solutions ensures high availability and redundan-
161
cy of other IT infrastructure components, such as SAN Fibre Channel

switches and especially servers. Fibre Channel switches can be made
redundant easily by installing another switch. The first path from the
server to a storage system goes through the primary Fibre Channel
switch; the second path goes through the secondary Fibre Channel
switch. With servers, the problem is more complex. Storage systems
and tape libraries ensure that we do not lose data stored on a storage
systems LUN. Except for a mapped LUN that is used for databases and
shared files, servers use internal hard drives to store operating system
and application installation instance. We can create a disk image that
will serve as a backup of this server drive, but then again, we will need
time to restore the copy, which increases the time for recovery. It is also
possible to load the operation system and applications over a LAN or
SAN from a storage system. However, this solution is not suitable for
all applications. There is always a time lag when the server hardware
failure occurs because we need to replace one physical server with another that is not broken. To prevent this lag, we need to make even
servers redundant. The technique that allows us to achieve this is called
server clustering.
Server clustering allows us to interconnect servers and their operations on the application level. There are computing clusters that combine the computing performance of all individual servers. The cluster
solution then provides the sum of the servers performance. However,
for the purposes of this book, we will be most interested in high-availability server clusters. You can see how such a cluster works in Figure
4.11.
162
External LAN
5
2
LAN 2
LAN 1
Heartbeat 1
Heartbeat 2
SAN 1
Server 1
Server 2
SAN 2
Figure 4.11 A failure resistant server cluster. Servers are connected to a storage system
through two Fibre Channel switches to meet redundancy requirements. The servers are also
interconnected by a LAN that allows communication with external users and between servers.
Notice the line between servers that is labeled as the heartbeat. The heartbeat is a clustering
software monitoring process. If Server 1 fails, Server 2 does not receive a heartbeat response
from Server 1, so it takes over the tasks Server 1 was working on.
To enable cluster functionality, you need to install a specialized clustering software layer on the servers. Servers then share the same tasks.
It is possible to implement an active-active cluster connection, which
provides load balancing. In this case, both servers are working together,
performing the tasks of one application. When we implement an activepassive cluster connection, only one server is processing the application
request and the second is in stand-by mode, ready to take over in case
the active server fails. The application status and configuration is syn-
163
chronized constantly. The clustering software allows you to set the conditions for processing takeover. Usually, the servers are communicating
their updated statuses in defined periods of time this is called the
heartbeat. If one server does not receive a response for a status update
request, it determines that the server on the other side is broken, and it
takes over immediately with no noticeable lag. This allows us to achieve
zero RPO and RTO.
The cluster protection can work within one location (organization
building). In this case, the IT infrastructure is completely resistant to
hardware failure and, to an extent, is resistant to power failure (depending on power source redundancy). It is also resistant to minor security incidents, such as theft and sabotage. To make your IT infrastructure resistant to natural disasters such as flood, fire or destruction of
the building, you need to geographically diversify your IT infrastructure.
This means that you need to replicate data to another location or create
geoclusters clusters over a WAN.
Remote replication: logical unit mirroring

Storage system based remote replication is the first step to geographically diversified IT infrastructure. It allows connecting a local disk
array to another storage system and to a remote storage system. The remote storage system can be located either in another building that belongs to the particular organization or in professional disaster recovery
data center. Remote replication is a nondisruptive, host-independent
solution for disaster recovery or data migration over any distance. The
key parameter that determines seamless remote replication is the type
of network connection to the secondary location. The connection can
be established using FCIP or iFCP protocols over the WAN (basically the
Internet), public line or dark fiber.
164
The public line that allows asynchronous transfer mode (ATM) represents a standard of unified network that serves both for transmission
of telecommunication services (landline) and data. It was designed to
achieve low latency and high speed. It requires a dedicated infrastructure. This technology is declining in favor of transmitting data over IP
networks.
Dark fiber is a specific technology based on dense wavelength-division multiplexing (DWDM). At your primary location, you have several
storage systems that need to replicate data over distance. Fibre Channel
cables from every storage system are plugged into a special multiplexing device that merges signals from multiple Fibre Channel cables into
one optical cable that connects both locations primary and secondary. At the secondary location, the multiplexing device transforms the
signal from one optical cable into more Fibre Channel cables connected
to each storage system. Dark fiber optical cable is capable of transmission over 100km. To enable data transmission over longer distances, it is
necessary to implement a signal amplification device a repeater. Dark
fiber technology requires an optical cable infrastructure that connects
both sites. Nowadays, it is possible to rent the dark fiber bandwidth
from a telecommunication company that owns optical wiring infrastructure. Dark fiber technology ensures high performance and reliability.
Once we have successfully interconnected storage systems at both
sites, we can perform standard replication procedures create, split
and resynchronize replication pairs. The only thing different from insystem replication is that we always have some network limitations,
because the bandwidth and reliability of local networking is always
higher than the bandwidth and reliability of solutions designed to carry
data over significant distances. To overcome these limitations, we need
to optimize remote replication. We need to consider network performance, especially latency, and decide whether we want to employ synchronous or asynchronous remote replication. Both solutions are highly
reliable, and their function will be further described.
165
DWDM/ATM/IP
any distance
Production
server
Write
acknowledge
1
Fibre
Channel
Extender
Fibre
Channel
3
Extender
S-VOL
P-VOL
te
Remote site
Primary site
Prima
Figure 4.12 Remote replication scheme. As you can see, DWDM, ATM or IP connections to a
remote site are possible.
Synchronous remote replication

Synchronous remote replication can be implemented when we have
a very good connection to the secondary site. It is especially important
to reach high transfer speeds and a very short response time (latency).
The key element that distinguishes synchronous remote replication
from asynchronous remote replication is write acknowledgement reporting. When a production server wants to perform a write operation
on a storage system, it needs to receive acknowledgement that the write
operation was successful. When we employ synchronous replication,
data is written to both storage systems at both sites simultaneously.
The primary site storage system needs to wait to receive acknowledgement from the secondary site. The primary site storage system can send
acknowledgement to the production server only if the second write is
successful. No sooner than when the data is written at both sites can
the server move on and request another write operation. This is why
we need such a good network connection to the remote site if the
latency is too high, performance of the production server is affected
166
because it has to wait for the remote site to send acknowledgement of

the successfully completed write operation.
The advantage of synchronous remote replication is obvious the
S-VOL located at the remote site contains exactly the same data as the
P-VOL at any point-in-time. This provides the smallest exposure to data
loss and fast disaster recovery. The high requirements on network performance and latency means that the distance between the primary
site and the remote site is somewhat limited. Synchronous replication
could hardly be employed between sites located in different countries
or even continents. Synchronous replication is likely to be implemented
if the remote site is located within the same city or region. The maximum theoretical distance for synchronous remote replication is 300km.
However, best practices recommend not implementing synchronous
replication if the distance between the sites is greater than 100km.
The strictly synchronous replication performs atomic write operations this means that the write operation either completes at both
sites or not at all. It also means that if connection to the remote site is
lost, the local system freezes and is unable to perform any operations
until the connection to the remote site is restored. This behavior guarantees zero data loss. However, it is usually preferred to continue write
operations to the local system if the connection to the remote site fails.
This provides continuous operation of the local site, but it can lead to
losing the desired RPO values.
Synchronous remote replication always affects the performance of

local production servers. The limitation is not technology, but rather the
speed of light. For a 20km distance, even when we use an Fibre Channel
connection, the roundtrip takes 134s, whereas nowadays cached in-system writes are completed in 10-20s. Deciding whether or not to employ
synchronous remote replication, therefore, requires careful consideration.
If the remote site is located in the same city as the primary site, the performance impact is usually not serious and allows synchronous remote
replication. The low distance between the sites, however, does not protect
our data from regional natural disasters such as earthquakes.
167
Asynchronous remote replication

In asynchronous remote replication, production servers do not wait
for write acknowledgement from the remote site. The write operation
is considered completed as soon as the local storage system sends a
write acknowledgement. Data from the local storage system is then
replicated to the remote site with a time lag. This enables long-distance
replication to a site located in another country or even another continent. This type of remote replication also provides the least impact to
host performance. The disadvantage is that data at the remote location
can lag a number of seconds behind the primary image. This means that
the most recent data can be lost if the primary locality fails completely.
When a production server requests to perform a write operation,
the local storage system sends a write acknowledgement to the host.
The host then considers the write operation successfully completed.
The local storage system then stores the differential data for a remote
site update into a log file. The log file is then sent to the remote site.
The remote site updates the S-VOL with data from the log.
Frequency with which the remote site is updated depends on network latency and bandwidth. The remote site can be updated nearly
instantaneously or in minutes, or even in hours in the most extreme
cases. Asynchronous data replication helps you ensure IT infrastructure
resistance to regional natural disasters. When implemented correctly,
asynchronous remote replication ensures that you lose a maximum of
a few seconds of the most recent data if the primary location fails completely.
Remote replication, both synchronous and asynchronous, is very
costly, and it represents the highend of data loss prevention solutions.
Another step that would ensure an even higher level of data protection
and data availability is to implement remote replication together with
geoclusters. Well discuss this step on the following pages.
168
The log of changes that is exchanged between sites during asynchronous replication is called a journal. Journal data is stored in a dedicated
journal volume (pool).
Dynamic replication appliance

Storage system based replication does not require the participation
of host servers; it merely mirrors the data stored on one LUN to another
LUN within the same storage system or to another storage system over
a distance. This solution is effective as long as you have homogenous IT
architecture (i.e., your infrastructure consists of storage systems from
one manufacturer). To manage disk array based replication, you would
use the software provided by the particular storage system manufacturer. If you use several storage systems from several storage system
manufacturers, then you also need to use several management tools to
implement replication. This is costly and inefficient. You need to unify
the replication management to simplify the procedures and save money that you would otherwise have to spend on extra staff, training and
management software licenses.
Essentially, you have two options:
Virtualization of storage systems you can buy an enterprise
level storage system with virtualization capabilities. This storage
system will virtualize all the storage systems you have, regardless of their manufacturers. You can then perform all administration procedures, including replication, with the software supplied
with this enterprise level storage system.
Replication appliance implementation you can also purchase
a dedicated device that takes care of replication. The replication
then does not take place on the storage system level, but on the
network level. The replication appliance, therefore, represents
network based replication.
169
A replication appliance is connected into a SAN, between hosts and

storage systems. The replication appliance then presents the storage array LUNs to the host. As all I/O operations pass through the appliance,
the write I/Os are detected and replicated to a storage system. The
configuration of the replication appliance is much more difficult and
complicated than storage system based replication because you need
to make adjustments for each server and LUN. It is also necessary to
consider the operations of the multipath software layer installed on the
server (two paths per server and LUN for redundancy and load balancing). This is called pass-through replication appliance implementation.
In this implementation, the replication appliance needs to provide sufficient performance when processing all I/O coming from servers so that
it does not slow down the entire SAN environment.
The replication appliance can be also implemented using agents running on production servers. The agents are small programs that split
write I/O into single copies. This happens on the production server. One
copy then goes directly to a mapped LUN on a storage system over the
SAN as if there were no appliance at all. In other words, the replication
appliance does not stand in the way of data; it doesnt even need to be
implemented in the SAN. The second copy is then sent to the replication appliance, which routes the data to the replica volume. The replica
volume is presented to the replication appliance either as a TCP/IP target (replication over LAN or remote replication over WAN) or an Fibre
Channel target (LUNs located in SAN require individual path for replication). These modes are called split write over TCP/IP and split write to
Fibre Channel target.
Implementation of a replication appliance can be complicated, but it
provides the best performance for large institutions with a heterogeneous IT infrastructure. The key features of this solution are:
Network based local or remote recovery and continuous backup
to disk
Continuous data protection (CDP) for optimal RPO/RTO ideal
for remote or branch offices (see Figure 4.13)
170
Rapid recovery
Support of disk-to-disk-to-tape storage for backup and activation
Single site or remote site data center support
Heterogeneous storage and server environment support
Application integration with Microsoft Exchange, Microsoft SQL
Server, Microsoft SharePoint, Oracle, etc. (dedicated agents designed to run on server)
The replication appliance is usually implemented in split write over
TCP/IP mode as depicted in Figure 4.13. Pass-through implementation
is possible but usually avoided because the replication appliance can
decrease the performance of data processing over the SAN.
Primary Data Center
Remote Data Center
Production Servers
Virtual/Physical Targets
LAN
Dynamic Replicator
Replicato
Appliance(s)
Figure 4.13 Possible implementation of replication appliance. Data is collected from servers
over the LAN and then sent to a storage system. Each server is running an agent that splits the
data.
Remote replication and geoclusters

For array-based remote replication, we need at least two storage systems, usually from the same manufacturer. One is installed locally, and
171
the other is installed in a professional data center or another building

that belongs to the organization. A local storage system is used by local
servers, while a remote storage system is used solely for replication purposes and can be installed without production servers. We know that
local high availability clusters add server redundancy that allows immediate data traffic takeover in the event of one server failure. Geoclusters
operate in a very similar fashion. The prerequisite for geoclusters is that
we have duplicated a fully redundant remote location. In this remote
location we not only have a storage system for remote data replication,
but also servers that are clustered together with the servers at the local
site.
Location A
Location B
WAN (TCP IP)
LAN A
LAN B
Interconnected LAN locations
SAN A
SAN B
WAN (FC)
FC Switch
2
1
Disk array A
Interconnected FC (SAN) locations
2
1
FC Switch
Disk array B
Figure 4.14 Geocluster interconnection scheme. Both sites (local and remote) are equipped
with the same nodes. Data from Disk Array A is synchronously replicated to Disk Array B over
iFCP, FCIP or dark fiber technology. Both SANs are interconnected. Servers in both locations are
also interconnected, usually using TCP/IP protocol.
Note that geoclustering, or long-distance clustering, as depicted in

Figure 4.14, looks quite simple. However, the whole problem gets more
complex when we have to mirror dozens or even hundreds of servers.
172
You can imagine that the configuration and installation is rather complicated. Crucial parameters that determine the overall performance of
geoclusters are again the network transfer rate and response time. Geoclustering with synchronous replication can be deployed to a remote
site located no more than 50km from the primary location.
When we talk about geoclustering, we are usually talking about
high availability clustering, not computing clustering, which would be
extremely demanding on network performance. The key words here
are cold site and hot site. A cold site is a remote site with hardware
that is powered down and consists only of the most important components. This solution is cheaper, but it takes time to resume all the
business processes if the primary site fails. For large institutions that
are highly dependent on data availability, hot site geoclustering is the
right solution. The whole IT infrastructure is mirrored. The heartbeat
between servers is checked regularly and specialized geoclustering
software is used. Complicated algorithms that are part of geoclustering software need to decide when to forward the traffic to the remote
location. In other words, it needs to evaluate whether the situation in
the primary location is critical enough to perform a swap operation.
A swap operation is usually performed by a tailored script that
switches the S-VOL status of the remote replica to the P-VOL and makes
it accessible for production servers. This script also restores database
consistency, usually by issuing a rollback command. Maintaining database consistency is of major importance and will be discussed later.
Three data center multi-target replication

Two data center replication strategies are viable for most in region
recovery for example, serving as a hot site for a campus level or metropolitan level server cluster or for out-of-region recovery sites where
propagation delays are not an issue. Synchronous replication provides
very fast recovery time (low RTO) and good data currency (low RPO).
173
However, asynchronous replication provides better protection against

regional disasters, albeit with a less favorable RPO. A three data center
strategy offers the best of both worlds: a combination of in-system and
remote replication that enables fast recovery and excellent data currency for local site failures, combined with advanced protection from
regional disasters.
Three data center multi-target replication therefore is the most advanced and secure solution that can be implemented to ensure uninterrupted operation of business processes. This solution can be extremely
expensive and it is suitable only for the organizations that are most sensitive to operation disruptions. See Figure 4.15.
Synchronous
Remote Copy
Primary Data Volumes

(Synchonous)
In-Region Recovery Data Center

Primary
Data Center
Ansynchronous
Remote Copy

(Ansynchonous)
JNL
JNL
Out-Region Recovery Data Center
JNL = Journal Volume

Primary JNL Data is stored before transmission.
Secondary/Remote JNL Correct record sequence is re-established after transmission.
Figure 4.15 Three data center multi-target replication. The maximum possible data protection is ensured by using two remote sites for data replication.
174
Replicated data consistency

In the first chapter, we described what data consistency is. For data
replication, data consistency is of major concern, especially when we
are replicating volumes that contain large databases. In this case, its
very important to maintain transactional consistency. Lets say an application is running on the production server. This application stores its
data in the form of a database. Items in the database are being updated
constantly; we are talking about live data. One update of one item is
one transaction. An application can use the database only when it is
consistent, (i.e., when there are no pending and incomplete transactions). In the context of storage systems, data (especially database) consistency is something that needs to be carefully considered when we
replicate the volume that contains a live database.
When we have a replication pair, the P-VOL is continuously synchronized with the S-VOL. During the synchronization, we cannot access
data on the S-VOL. When we need to access the data on the S-VOL,
we need to split the pair. If we split the pair without considering data
consistency, it can easily happen that data on the S-VOL will contain
incomplete transactions and thus be unusable for an application and
for testing or recovery. That is why we either need to stop the database
operations (transaction update on the P-VOL) on the production server
before splitting the replication pair, or we need to use an agent installed
on the production server that communicates with the storage system.
When the agent determines the database is consistent, it allows the
storage system to perform pair split operations. Then both the P-VOL
and S-VOL are consistent and can be accessed by an application. Note
that it is the data on the P-VOL that needs to be consistent before the
pair split. This does not necessarily mean that we need to completely
stop the production server. Transactions can be still handled they are
just not stored to the P-VOL during the time of pair split operation; they
are stored in the storage systems cache. Once the pair split operation is
complete, data from the cache can be flushed to the P-VOL again.
175
When we replicate data with asynchronous replication to a remote

location, we could lose a few transactions or parts of transactions because of synchronization delay. The remote replica then contains corrupted and inconsistent data. An application that uses the particular
database then cannot work with the database until the consistency is
somehow restored. Consistency is usually restored by issuing a rollback
command (SQL), which restores the last known consistent state of the
database by erasing the incomplete transactions.
Continuous data protection (CDP)

The term continuous data protection denotes a data loss prevention
strategy that is different from conventional backup and data replication
strategies in several ways. Continuous data protection is sometimes also
called continuous backup or real time backup. Continuous data protection is usually provided as a functionality of advanced backup software
that allows backup of every single change made to data.
In a traditional backup, we need to determine backup windows and
backup scheduling. We can schedule backup once a day or even every
hour. If there is a problem, we can restore the system to the previous
state as it was captured during the last backup. Unlike traditional backup, no scheduling is necessary in CDP, simply because the procedure is
ongoing. Every change in data is traced and backed up.
There also is a difference between data replication and continuous
data protection. Data replication only protects one copy of data, the
most recent, because it mirrors the current state of the P-VOL. If the
data stored on the P-VOL becomes corrupted or is purposefully deleted,
these changes will be automatically propagated to the secondary volume without the possibility of restoring the data to the previous state
before the data corruption or deletion occurred. CDP, on the other hand,
allows you to restore any version of data at any point of time. You can go
176
back step by step, reversing all individual changes that have been made.
Continuous data protection works on the block level, not the file level.
This means that if we have a Microsoft Word document and you change
one letter, the backup software does not save a copy of the whole file.
It saves only the changed bytes. Traditional backup usually make copies
of entire files. The CDP technique is therefore more disk space efficient.
The downside is that continuous data protection can be expensive to
implement, and it can also heavily affect LAN performance, especially if
the data changes are too frequent and extensive. CDP is, therefore, not
very suitable for backup of large multimedia files. Small changes in databases or documents can be handled efficiently because we are talking
about relatively small amounts of data. In the case of multimedia processing, changes in data can be too large (gigabytes of data). The whole
backup procedure is then slow, and LAN performance is highly affected.
If the CDP functionality is supported by your backup software, it can
usually be implemented using your existing backup infrastructure. CDP
functionality requires the presence of agents on the production servers.
Redundancy in basic infrastructure

The concept of redundancy has been accompanying us from the very
beginning of this book. Data management as a whole is mainly about
three things: what data we need to keep, where we keep it and how
can we prevent its loss. Business continuity management adds requirements on high availability of data that is vitally important for operation
of critical business processes. This all leads us to fully redundant architecture. We need to have redundant copies of data through RAID implementation, data replication, data backup and data archiving. To ensure
these techniques work seamlessly, we need redundant hardware: dual
controller architecture in a storage system, redundant SAN components
(switches, HBAs, NICs) and also redundant servers (clusters). We have
177
even discussed location redundancy issues: remote replication and geoclustering. The trick is that you need to have everything at least twice.
The last thing to take care of is a redundant basic infrastructure.
By basic infrastructure we mean especially power supply, air conditioning of server rooms and Internet (WAN) connectivity. When there is
a power outage, a storage system powers down immediately. Installed
batteries keep data in cache alive for at least 24hrs, which allows operation resumption with no data loss once we have power back on. In
some storage systems, large batteries are installed that allow flushing
the data from the cache to the designated hard drive. Once this operation is completed, the whole system powers down, including the cache.
To overcome short power outages that last for seconds or minutes,
we can install uninterrupted power supply (UPS) units. A UPS device
is equipped with extremely powerful batteries that are able to provide
electricity to servers usually for several minutes. There is a large scale
of UPS products that can be used with home computers as well as with
whole server rooms. UPS devices able to support whole server rooms
are very expensive. If a power outage occurs, the UPS provides electricity stored in batteries. UPS batteries can last for several minutes, which
is enough to overcome short power outages or cover the time necessary
for switching to an alternate source of electricity, such as a power generator. If the UPS batteries are nearly depleted and the electricity supply has
not been resumed, the UPS unit starts a controlled shutdown procedure,
safely shutting down all servers, preventing a cold/hard shutdown. Large
companies and especially data centers have redundant power sources
such as connection to two separate grid circuits or a diesel engine based
power generator.
IT equipment produces a lot of heat. That is why server rooms are
equipped with air conditioning. If there is a disruption of air conditioning, the threat of IT equipment overheating becomes imminent and all
the servers and disk arrays must be powered down to prevent hardware
damage. That is why air conditioning must also be made redundant.
Regular air conditioning maintenance is necessary as well.
178
Internet connectivity and access to public data circuits based on optical cables are another threat of operation disruption. An organization
should ensure connectivity availability by contracting at least two connectivity providers.
Tests and drills of disaster recovery procedures

Business priorities of organizations are changing rapidly as new technologies emerge and the market demands these new technologies.
That is why any business continuity documentation needs regular maintenance to reflect these changes and to ensure currency of the plans.
A disaster recovery (DR) plan is part of a business continuity plan that
deals with securing IT infrastructure operations and preventing data
loss. To make sure the disaster recovery plan is effective, extensive testing and drills must be conducted.
DR plan efficiency can be tested in a virtualized testing environment
that simulates real applications and equipment the company uses for
its business. A virtualized testing environment is a very safe way to test
the efficiency of disaster recovery plans, as it is completely risk free.
However, it can be difficult to simulate all the processes in a virtualized environment. That is why testing is also conducted with production
equipment.
Every particular threat that is dealt with in the DR plan can be
simulated. The drill then involves all members of the DR team and IT
equipment necessary for operation of the particular process that can
by disrupted by the threat. For example, we can conduct a drill based
on the scenario that the organizations building has been destroyed or
rendered inaccessible. The DR team then activates the DR plan as if the
situation were real. Personnel then proceed with the plans individual
steps meeting at emergency location (offices), restoring data from
the remote replica, switching the traffic to a remote location geocluster,
179
etc. These drills are very effective and help to discover weaknesses of
the documentation. Tests and drills are an essential part of business
continuity documentation, and their importance should not be underestimated.
Disaster treatment overview

In this chapter we have mentioned the most important techniques
of securing data and preventing its loss. Before we close this chapter
and move to virtualization concepts, we can review the most common
threats an organization can face and suggest relevant technical solutions. Remember that none of the techniques mentioned in this chapter will save us on their own. To design appropriate solutions, we need
to employ a combination of functionalities tailor-made to measure an
organizations particular infrastructure.
180
Disaster
Power cut
Stolen
equipment
Fire
incident,
minor
RTO
Disaster Impact Prevention
Delayed
RTO
UPS, storage system equipped with batteries to

support cache
Immediate
RTO
Power generator, redundancy connection to

different power grid circuit
Delayed
RTO
Disaster recovery plan activation, ordering new

replacement equipment, local server cluster
data recovery from onsite or offsite tapes
Immediate
RTO
Data recovery from remote replication site, hot

site geoclusters, CDP
Delayed
RTO
Disaster recovery plan activation, ordering new

replacement equipment, data recovery from
offsite tapes
Immediate
RTO
Delayed
RTO
Flood
Local server cluster, local replication, local

tapes, remote replication, geoclusters, CDP
Emergency location (offices), remote cold site,
ordering new replacement procedure, offsite
tapes
Immediate
RTO
Geocluster solution, synchronous replication
Delayed
RTO
Emergency location in another region, offsite

tapes, asynchronous remote replication to
distant datacenter
Immediate
RTO
Geoclustering, three datacenter multi-target

replication
Earthquake
Figure 4.16 Disaster treatment overview table.
181
Completely Duplicated/
Interconnected Hot Site
Remote Disk Mirroring
More
Disk Mirroring
Shared Disk
Disk Consolidation
Single Disk Copy
Electronic Vaulting
Tape Onsite
Tape Backup
Offsite (trucks)
Importance
of Data
More
Amount
of Data
Less
Less
Delayed
Recovery
Time
Immediate
Figure 4.17 Diversity in data protection requirements.
All data is not created equal. It is likely that only a portion of data is
critical to the basic operation of a company. The key is to think through
the data protection requirements for different classes of data. Its quite
likely that in most scenarios you will have some subset of data that
would warrant remote disk mirroring. Refer to Figure 4.17 to see appropriate solutions depending on the data importance, recovery time
objective and amount of data.
Fibre Channel offers the best performance for both short and long
distances, but it has the highest cost. NAS and iSCSI are at the low end
of the cost spectrum, but they sacrifice performance for both short and
long distances. FCIP, iFCP and DWDM are solutions that perform better than NAS and iSCSI but at a higher cost. However, they are cheaper
when compared to Fibre Channel. See Figure 4.18 for reference.
182
High
FC
DWDM
NAS, iSCSI
Cost
FCIP, iFCP
Longest
Low
Distance
Less
Worst
Performance
Best
Figure 4.18 Solution area of cost, performance and distance.
This chapter provided you with the most important aspects of business
continuity planning and solutions that enable disaster recovery. Remember that the final design of failure resistant IT infrastructure is always a
combination of the techniques mentioned, and it must reflect the needs
of the particular organization.
Virtualization of storage
systems

How to describe virtualization and its benefits
What layers of virtualization there are and what is possible to
virtualize
How we can virtualize servers and why we do it
What virtualization possibilities we have in storage systems
The difference between fat and thin provisioning
Virtualization concepts in a SAN
184
5 Virtualization of storage systems
What is virtualization?
When we take a look at trends in information technology, we can observe several important facts. For example, we can agree that hardware
is getting more and more powerful. You have probably heard about
Moores law, which says that every two years the processing of computers is doubled. At the same time, the price of hardware is decreasing. It
means that, for less money, you can get more performance. This is more
or less applicable in every area of IT. Powerful enterprise level storage
systems used to be so expensive that only a few organizations could afford them. Nowadays, a midsized organization can afford to purchase
IT equipment that was too expensive for government institutions and
banks about ten years ago. As hardware gets more and more powerful
and costs less and less, we can observe another phenomenon software becomes the most expensive item on the shopping list.
In the past few years the hardware became so powerful that it was
nearly impossible to achieve good utilization of all the performance it
could provide. The traditional model is based on a one server per one
application scheme. (Refer to Figure 5.1.) If your company uses 10 applications, then you need to have 10 physical servers. Lets say that out
of these 10 applications, only three are mission critical and can use all
the performance of a server. The remaining applications are not used
so intensively and thus require only about 25% of the actual computing
power of a server. Yet all 10 servers consume power, have high cooling
requirements, occupy some space and need to be backed up or clustered. This leads to inefficiency because you have 10 servers and you
pay for their operations, but you actually use less than 50% of their
computing capacity.
Virtualization addresses this problem. If you looked up the definition of the word virtualization you would probably find that virtualization is an abstraction of hardware. This is true, but it can be hard to
imagine what actually lies behind this definition. Virtualization takes all
the hardware resources, such as processor performance, RAM size and
disk space, and creates aresource pool. This resource pool can then be
185
sliced into several logical a resources, which we call virtual machines. If

you virtualize a server, you get several logical servers. Each of these logical servers can then be used for operating system installation and for
applications. The operating system works on a virtual machine without
knowing it is a virtual machine; the operating system does not communicate directly with hardware but with the virtualization layer, which
provides an interface between the hardware and a virtual machine. The
virtualization layer allows you to configure how much physical resources the virtual machine can use. Virtualization is achieved through implementation of specialized software. Alternatively, it can be provided as
an inbuilt capability of a particular device (such as a storage system). If
the virtualization functionality is inbuilt, configuration is made through
a microcode interface, which can be either a CLI or GUI.
If we get back to our 10 applications and 10 servers example, we can
see that with virtualization we need only one or two powerful servers
to run 10 virtual machines, each with its own instance of operating system. This allows you to save money and achieve good utilization of all
the physical hardware you buy.
AP
P
OS
AP
Ap
pli
Op
cat
era
AP
OS
OS
Vir
tu
ion
aliz
tin
gS
yst
e
Traditional Architecture
ati
on
AP
AP
P
OS
AP
P
OS
P
OS
Lay
er
Virtual Architecture
Figure 5.1 Server virtualization. The traditional architecture model requires one physical server
per operating system and application. A virtualized server is able to run several virtual machines
that all share the physical hardware.
186
Virtualization is the abstraction of computer resources. It hides the

physical characteristics of computing resources from the way other systems, applications or end users interact with those resources. Once virtualization takes place, a single physical resource appears to function as
multiple logical resources. It is also possible to virtualize multiple physical
resources so that they appear as a single logical resource. In personal computers and servers, logical resources are called virtual machines.
Virtualization: area of application

Virtualization can be implemented at least to some extent in all devices used in information technology. Nowadays, virtualization is often
employed in:
Personal computers
Servers
Storage systems and their components
If you do not have much experience with servers and storage systems, you can take a look at several examples well known from personal
computers. We have already mentioned hard drive partitioning. If you
have a single hard drive installed in your computer, you can divide it into
two logical units. In a Microsoft Windows environment, they appear
as C: and D: volumes. Both these logical volumes can be accessed
and used as if they were two separate physical hard drives installed in
your computer. Disk partitioning or disk concatenation can be seen as
virtualization because the operating system is accessing the hard drive
through the two logical units. Widely used virtual CD/DVD drives are
another example of virtualization in personal computers. A specialized
software will create a virtual CD/DVD drive, which is able to open images (or clones) of actual CD or DVD medium. Virtual CD/DVD drives
are very popular because there are lot of CD/DVD images available on
187
the Internet. Usage of virtual drive eliminates the need to burn these
images on physical media before usage. CD/DVD imaging is also often
used for backup of purchased film DVDs or audio CDs.
Some operating systems also natively allow you to run another operating system. We can take Windows 7 Ultimate as an example. This
operating system allows you to run older Windows XP as a virtual machine to enhance compatibility of user application with operating system. If you have an old application that does not run in a Windows
7 environment, you can set up a virtualized Windows XP environment
in which the application runs without problems. It is also possible to
purchase specialized virtualization software, such as VMware Workstation, that allows you to run several instances of an operating system.
This can be useful for setting up a testing environment. Information
technology students, software developers or IT administrators can set
up several virtual machines while running an ordinary computer with
regular client version of operating system. There they can perform application development or testing without the interference with other
installed applications and operating systems. VMware Workstation is installed within the operating system. Thanks to encapsulation, all virtual
machines are then saved on the hard drive in the form of files that can
be copied or easily transferred to another computer. Virtual machines
are then accessed in a similar fashion as remote desktops you can
open a full screen window with an OS instance.
VMware Workstation and similar virtualization applications usually
require a CPU that natively supports virtualization. Nowadays, most
processors in personal computers offer virtualization capabilities. Hardware support of virtualization is sometimes referred to as virtualization
acceleration. Virtualization support in a CPU allows virtual machines
to effectively access hardware means through the virtualization layer
without performance overhead. This leads to very good virtual machine
performance.
In servers, virtualization is becoming a standard for several reasons.
Virtualization in servers allows you to achieve a higher level of hardware
188
resource utilization. Virtual machines are also much easier to manage

because of unified and consolidated environment. If you have several
physical servers, they are likely to contain different hardware components. This requires additional configuration of different device drivers.
In virtual machines, you always have a unified environment. The operating systems see the virtualization layer as a set of physical components.
This means that if you open the operating systems device manager, you
can see virtual processors, virtual RAM, virtual network interface cards
and virtual host bus adaptors as if they were real. The virtualization
layer takes care of linking these virtual components to their physical
counterparts. Virtual components are the same on all physical servers.
Therefore, the manageability of servers is increased you do not have
to take care of updating different components with different drivers.
Virtual machines use encapsulation. Each virtual machine, with all
its components operating system, applications, user data, etc. is
represented by a single file. This allows very simple migration. You can
actually copy a virtual machine on an external hard drive and run it on
another server without any modifications. Virtualization software that
runs on servers is more sophisticated than virtualization software designed for personal computers and is mainly for testing purposes. It must
provide enhanced capabilities for virtual machine backup, migration, redundancy, high availability and high performance. The possibilities we
have with server virtualization will be described later in this chapter.
Virtual Machines
Application
Application
Application
Application
Windows
Server
2008 R2
Windows
Server
2008 R2
Linux
Solaris
Virtualization Layer
Physical Hardware
CPU
RAM
Disk
LAN
HBA
Figure 5.2 The virtualization layer provides interface between physically installed components
and virtual machines. Virtual machines see unified
virtual components that
are linked to their physical counterparts.
189
Server virtualization has been implemented for quite some time

and it is becoming standard in most larger institutions. Storage system virtualization, on the other hand, is more complex. The very basic
virtualization technique that takes place in a storage system is RAID. A
single parity group consists of several hard drives linked together by
RAID functionality to provide a logical storage area, which can then be
further partitioned. RAID is implemented for redundancy and performance reasons because data striping allows us to achieve faster read
and write transfer rates. Every storage system offers RAID functionality. RAID, similarly to partitioning, represents virtualization of physical
hardware resources. However, inclusion of RAID functionality in a disk
array does not qualify as a virtual storage system.
When we talk about the virtualization of a storage system, we usually do not mean RAID functionality because it is somehow considered
as a given. We typically mean techniques that allow virtualizing a single
storage system by partitioning its resources, such as cache, ports and
hard drives. One storage system then appears as two or more. Other
virtualization capabilities make it possible to virtualize several storage
systems by various vendors. A storage system equipped with such virtualization capabilities consolidates other storage systems and unifies
their management. Among other virtualization capabilities of a storage
system are pooling and functionalities based on it, such as thin provisioning. In pooling, we put several parity groups together and we create
a pool that can be further used for thin provisioning. Thin provisioning
allows us to allocate more storage capacity to the host than is actually
physically installed on a storage system. Well describe how this is done
and why we do this later in this chapter.
Storage based virtualization is always managed in the storage system without the participation of a host. The functionality is included
in the storage systems microcode and is usually subject to additional
licensing. In other words, you have to pay to enable these functionalities. Storage system virtualization is configured directly in the storage
systems microcode using a CLI or GUI.
190
Layers of virtualization
We can virtualize personal computers, servers, storage systems and
even networks. We can picture this as horizontal virtualization. However virtualization can also be vertical, implemented in layers. When we
have a storage system, we get the first layer of virtualization when we
create a parity group, or RAID array. When we take several parity groups
and create a storage pool, we take already virtualized logical storage
and virtualize it even further. Out of this pool we can create several virtual volumes that will be dynamically allocated the storage space; this
is yet another layer of virtualization. We usually deal with several layers
of virtualization, which are implemented to provide enhanced functionality. The host will then access the highest layer of virtualization in
this case a virtual volume without knowing about other virtualization
layers. Only virtualization software knows where the data is physically
located, because data is striped within a single parity group across several physical hard drives and then over several parity groups within one
storage pool.
Users
Virtualization
Application
Virtualization
Compute
Virtualization
Storage and Data
Layers of virtualization can also

be seen when we look at our IT infrastructure as a whole the host
is accessing a virtualized storage
area, either a LUN or, in the case of
thin provisioning, a virtual volume.
The server can also be virtualized.
Applications can run on several virtual machines that share one physical server. Users can then access
the application from yet another
virtual environment, such as a virtual desktop.
Figure 5.3 Virtualization layers in IT infrastructure.
191
Virtualization benefits
Lets review a summary of the virtualization benefits we have mentioned in this chapter. The most important motivation that leads us to
virtualization implementation is management simplification. Virtualization is implemented to simplify management of heterogeneous IT infrastructure. The individual benefits of virtualization are:
Migration virtual machines and LUNs can be easily transferred
from one physical device to another. In some cases this is possible
without disruption of operation. This means that we can transfer
a running virtual machine with OS and applications to another
hardware environment without impacting its operations. Users can still access the application during the migration. This is
achieved by encapsulation of a virtual machine into a single file.
Backup encapsulation also makes it easier to back up the
whole virtual machine. Virtualization software usually provides
backup functionalities.
Hardware platform independence physical servers can be of
different configurations, yet the virtualization layer ensures unified environment for virtual machines. This means that within
one virtualization platform, you can transfer virtual machines
from one physical server to another regardless of their physical
configuration. There are no issues with drivers.
Enhanced utilization virtualization allows effective use of resources. A virtual machine is allocated the exact amount of computing performance, RAM and storage space it requires, and
no more. If it requires more, other resources are provided dynamically and nondisruptively. Imagine you are running a virtual
server that is under high load. The high load is detected by the
virtualization platform and the virtual server is allocated another
processor that would support its performance.
Unification unification of infrastructure allows you to save
money. You can manage several storage systems by different ven-
192
dors with only one piece of software. Functionalities that require

additional licensing are enabled (purchased) only for the storage
system that takes care of virtualization. The functionalities enabled in the virtualizer are automatically propagated to another
storage system. Imagine you want to employ remote replication.
Without virtualization, you would have to purchase a license for
each storage system, which is costly. With virtualization, you purchase a license only for the virtualizing storage system. Remote
replication is then available for all storage systems.
Virtualization enables you to run several virtual machines on one
physical server. This means you do not need to invest too much in
physical equipment. This leads to lower consumption of electricity and lower cooling requirements.
Virtualization helps you achieve lower RPO and RTO. It is much
easier to conduct disaster recovery procedures in a virtual environment, especially thanks to unification, encapsulation and
detachment from physical hardware. The virtualization layer can
run on several servers. When one server fails, other physical servers provide sufficient means to support all virtual machines.
Physical resources can be added without disruption. You can
add memory, processors or even whole servers to increase performance of virtual machines without actually disrupting their
operations.
VMware based server virtualization

VMware is one of the best known providers of virtualization solutions
for servers. We have already mentioned VMware Workstation, which is
well suited for use in personal computers, mainly for testing purposes
and application development. For server virtualization, VMware offers
two essential products VMware Server and VMware ESX Server.
193
VMware Server is installed into an existing operating system in the

same manner as any other application. The operating system can be
either Microsoft Windows Server or a server version of Linux. VMware
Server then runs as operating systems services and provides tools and
management for virtualized environment. When you start the control
console you can immediately begin to create and configure virtual machines. Each virtual machine is allocated virtual resources, such as processors, memory, hard drives, network interface cards and graphical
adaptors. Once you have your virtual machines configured, you may
proceed with installation of individual operating systems. These operating systems run independently and can be different types. At the end
you have several virtual machines with instances of operating systems
running on a single physical server. In addition, the originally installed
operating system with VMware Server is still available and running together with virtual machines. This type of solution is very easy to implement and provides a good level of performance and reliability. However, the limitation is that VMware Server does not have direct access to
hardware means; it can only use resources provided by the underlying
operating system. VMware Server has to communicate with hardware
through an interface offered by the originally installed operating system. This does not have to be an issue, but you need to accept at least
slightly diminished virtual machine performance. A problem in the original instance of the operating system can result in instability of all virtual
machines. The maximum possible stability of a virtualized environment
and the best usage of hardware resources can be delivered when you
use another VMware product VMware ESX server.
In VMware ESX implementation, you do not need any underlying,
previously installed, operating system. ESX Server contains its own dedicated and thin operating system and can therefore be installed directly
on a physical server. This ensures direct access of the virtualization layer
to hardware resources, delivering the best possible performance. Once
you have ESX Server installed, management and configuration of virtual
machines is practically the same as in VMware Server. The major advantage of ESX Server is stability and performance. In addition, ESX Server
provides a range of advanced functionalities for efficient clustering,
194
backup, disaster recovery and virtual machine migration. In ESX Server,

a virtual machine can be transferred nondisruptively to another physical server. If we upgrade our hardware or simply purchase a new server,
existing virtual machines can be easily transferred to this new server
while on and running. Virtual machines can be easily updated, too. You
can allocate processors, memory and other virtual components without
the need to reboot the virtual machine. You can also configure these
resources to be allocated automatically when needed.
Thanks to its flexibility and performance, VMware ESX Server is well
suited for implementation in enterprises.
Microsoft Hyper-V based server virtualization

Hyper-V is a virtualization solution developed by Microsoft. It is included in the Microsoft Windows 2008 Server operating system, but it
can also be purchased as a standalone product called Microsoft Hyper-V
Server.
In Windows 2008 Server, you can enable Hyper-V as a server role.
This means that you need to have an installed instance of Windows
2008 Server, and virtual machines are then configured similarly as in
VMware Server. If you choose a Hyper-V Server role, other functionalities of Windows 2008 Server can be limited and certain services may be
unavailable. Through integration with an operating system, you achieve
high performance and reliability.
Hyper-V is a hypervisor based virtualization solution. Hypervisor is
another, more specific, name for the virtualization layer. The Windows
2008 Server instance is installed as a parent partition. Virtual machines
are then located in child partitions. Operating systems installed on virtual machines are presented with virtual resources. These virtual resources are linked to hardware resources through VM buses (VM stands for
195
virtual machine). Here we can see a difference between VMware Server

and Hyper-V. VMware Server can access only resources presented by
the operating system. VM buses are native to Windows 2008 Server and
they facilitate direct access to hardware means in Hyper-V based virtualization. The Hyper-V Server solution then works similarly to VMware
ESX Server neither of them requires an underlying operating system
instance.
Child Partition
VMI Provider
Virtual Machine
Management Service
Applications
Applications
VM Worker
Processes
Windows
Kernel
Device
Drivers
Virtualization
Service
Provider
(VPS)
VMBus
Virtualization
Service
Consumer
(VSC)
Windows
Kernel
VMBus
Virtualization
Service
Consumer
(VSC)
VMBus
Windows
Kernel
User Mode
Child Partition
Kernel Mode
Parent Partition
Hypervisior
Hardware
Figure 5.4 Server virtualization with Hyper-V. The hypervisor virtualization layer is thin and
optimized for direct access to hardware resources. Virtual machines in child partitions access
the virtualization layer through the VMBus interface. Device drivers for virtual machines are
loaded from the parent partition with the original instance of Windows 2008 Server. Virtual machine configuration and management are also done in the parent partition operating system.
Virtualization with blade servers

Before we get to advantages of blade server virtualization, perhaps
we should consolidate our server knowledge. A server is a computer
that provides services to other computers on a network. Basically, any
desktop computer or even notebook can assume the role of a server.
196
Due to high requirements on reliability, availability and performance,

we use computers that differ from those you have at your home or office. The computers designed to work as servers are equipped with high
performance processors, more durable and reliable hard drives, motherboards that can bear significantly more RAM modules and specialized
interface cards such as HBA for Fibre Channel connection. A server computer can look exactly the same as any other personal computer; it can
be a rack server or it can be a very modern blade server.
Computers that are modified so they can be installed in a rack are
called rack servers. These are the most common. A rack is a standardized chassis that can be equipped with many IT components such as
rack servers, switches, hubs, tape libraries and midrange storage systems. Rack servers are more space efficient than standalone servers.
The most modern type of server design is the blade server. A blade
server has a specialized chassis that provides shared powering, cooling, storage and networking resources. Individual servers then look like
computer cards that are simply plugged into the chassis. A blade servers chassis is then installed into a standard rack. Individual blades then
contain processors and memory modules. Blade servers represent a
highly scalable solution that saves a lot of rack space. Blade servers also
provide simplified cable management. Aside from these characteristics, blade servers provide additional functionalities that can differ from
vendor to vendor. Because of their robustness and complexity, blade
servers are well suited for clustering and virtualization. Some vendors
provide built-in virtualization capabilities based on logical partitioning of the installed hardware resources. This functionality is similar to
mainframe management, where you have one super computer further
partitioned into logical computing units.
The logical partitioning of hardware resources is a functionality available in microcode stored in components that are part of a blade servers
chassis. Through the management console, you can create logical servers (basically virtual machines) and allocate them directly with physical
resources. The number of physical resources depends on the number of
197
blades with CPUs and memory installed in the chassis. Individual blades
are hot swappable. This is a powerful solution that does not require the
participation of any third party virtualization software. Logical partitioning functionality sometimes hides behind the abbreviation LPAR.
Figure 5.5 Blade servers are installed in a blade server chassis. The chassis is then placed in
a standard rack. In this picture we see Hitachi Data Systems blade servers that are an essential
part of Hitachi Content Platform. These blade servers offer logical partitioning, which is a highly
sophisticated form of server virtualization.
VM hardware requirements and networking

All virtualization strategies based either on third party virtualization
software or virtualization capabilities built-in microcode allocate virtual
machines with computing and memory resources. The virtualization
implementer or administrator has to decide how many physical resources will be allocated to a particular virtual machine. The minimum
198
hardware requirements are usually determined by the operating system requirements to be installed on the virtual machine. The most important resources are computing power represented by the number of
processors or processor cores and the size of RAM. For small dedicated
servers, we usually need at least one virtual processor and 2GB of RAM.
With robust solutions and servers that run mission critical applications
under high load, we have up to 64 processor cores and 256GB of RAM
per virtual machine. These are rough numbers to give you a general
idea of virtual machine hardware requirements.
A standalone virtual machine would hardly be of any use with no
connection to a network. Virtual machine networking is configured from
a virtualization software management console by allocating a network
interface card (NIC) and HBA cards. Once you add an NIC, you may configure it in the traditional way using common protocols, usually meaning TCP/IP protocol for LAN networks and Fibre Chanel Protocol for SAN
networks. When we connect a virtual machine to physical network, we
say we bridge the connection. Virtualization software also provides the
functionalities of a virtual switch by enabling mutual communication
among virtual machines.
Memory overcommitment in server virtualization

When we create a virtual machine, we allocate it with hardware resources. Common sense would suggest that you can allocate only as
much hardware resources to virtual machines as you have physically installed in a server. To achieve the best possible utilization of your hardware, you can, in certain situation, allocate more hardware resources to
virtual machines than you have physically installed in your server. We
are talking about RAM in particular. This can be especially helpful when
you have several virtual machines that are not under high load and are
not showing good utilization. One example is desktop virtualization. In
desktop virtualization, you have a server that is running dozens or even
199
hundreds of client operating systems instances. The reason for this approach is to use one powerful server with high computing performance
instead of dozens of expensive and powerful client computers or laptops. Employees of the company then use thin clients, for example, less
powerful, inexpensive computers that serve only for connection to a
virtualized client operating system running on a server. This is similar
to a remote desktop connection, only the performance and stability of
this solution is much higher. Desktop virtualization also simplifies management and backup of client workstations. The example of memory
overcommitment is based on two assumptions. The first is that not all
employees are using their computers simultaneously. In other words,
there is always a certain number of virtual machines that are turned off.
The second assumption is that the client operating system is not using
the allocated memory completely. If you cleverly allocate more hardware resources than physically available, you save money and increase
utilization, because you can install more virtual machines per server.
Memory overcommitment is not limited to desktop virtualization; there
are other areas of application. The IT administrator must decide how
much memory to overcommit. If the overcommitment were too high,
running virtual machines could potentially use all the physically installed
memory, and on the attempt to use memory beyond what is physically
available, the virtual machines would crash.
Memory overcommitment is another example of a virtualization
technique. Even though it may seem illogical and potentially dangerous, it can increase utilization of hardware and save costs when skillfully
implemented. In the context of storage systems, memory overcommitment is similar to thin provisioning.
Thin provisioning
Provisioning is the process of allocating storage resources and assigning storage capacity for an application, usually in the form of server
200
disk drive space. The storage administrator has to determine the applications need for storage and allocate adequate resources on a storage
system. These resources are then mapped to a server as a logical unit
(LUN) and made available for an application. Nowadays, we differentiate between the traditional model of provisioning (also called fat provisioning) and thin provisioning.
When you deploy a new application, you need to assign it a storage
area. To avoid future service interruption, it is common to overallocate
storage by 75% or more. This is called fat provisioning. Imagine that
your application requires a storage area of 100GB. You expect that application requirements on storage will rise in future. To ensure that the
application always has enough storage, you allocate 200GB. To allocate
200GB you need to have this capacity installed in your storage system.
If you do not, you need to purchase additional hard drives. Remember
that with RAID implementation and possible employment of in-system
replication you may need about 1TB of physically installed capacity to
be able to provide 200GB of storage for this particular application. This
can be costly, especially if you have dozens of applications. To achieve
higher utilization, you can implement thin provisioning functionality.
In thin provisioning, whenever there is a new application, its entire
storage allocation is not required initially, so the application is allocated
virtual storage. This is similar to memory overcommitment in server virtualization.
Purchased,
Allocated
BUT UNUSED
Actual DATA
Fat Provisioning
Purchase
capacity
As needed
Initial
purchase
and
allocation
Purchased, Allocated
BUT UNUSED
Alert message when

additional storage
required
What you
need initially
Purchased, Allocated
BUT UNUSED
Initial
purchase
Actual DATA
Thin Provisioning
Figure 5.6 Comparison of fat and thin provisioning.
201
To enable thin provisioning, you need to set up parity groups that

are configured in the desired RAID level. Several parity groups are then
virtualized into one storage pool. The servers and their applications are
then allocated virtual volumes. Virtual volumes themselves do not represent any physically installed capacity. Virtual volumes are mapped as
LUNs, and servers cannot see the difference between a traditional LUN
and a LUN that hides a virtual volume. Virtual volume size can be configured without concerns about the physically installed capacity. When
a server needs to write data on an allocated virtual volume, storage is
automatically provided from the storage pool. Similar to copy-on-write
snapshots, a virtual volume contains only pointers that point to the actual location of data in a storage pool.
SERVERS
DP Volumes (Virtual Volumes)
HDP POOL
Parity Groups/HDDs
Figure 5.7 Thin provisioning. Parity groups are added to a thin provisioning pool. Virtual volumes are mapped to servers. Virtual volumes do not contain any actual data. Data is stored in
the storage pool. Virtual volumes contain pointers that point to the location of data in the pool.
202
Thin provisioning is a modern trend that reaches both enterprise

level storage systems and midrange storage systems. It is the most common virtualization technique used in storage systems.
Take a look at the list of thin provisioning benefits:
Storage performance
Storage pools allow increased physical disk striping. Data is
striped within RAID implementation and then again over parity
groups. This helps to avoid bottlenecks.
Increased physical disk striping evenly spreads I/O activity across
all available physical disk resources.
Thin provisioning reduces the need for performance expertise
and analysis. It is simple, and thanks to added performance, you
may not need to employ complicated performance tuning.
Storage administration
Thin provisioning simplifies application storage capacity planning
and administration.
It increases storage utilization and reduces wasted storage capacity.
Thin provisioning eliminates downtime for application storage
capacity expansion.
It improves application uptime and SLAs.
We can see that thin provisioning not only saves money, but it also
delivers increased performance and simplified management. In traditional (fat) provisioning, you need to determine what requirements the
particular application will have in the future, and you overallocate the
storage using physically installed capacity. If you happen to run out of
space on an allocated volume, you need to extend the mapped LUN.
This is fairly simple from the perspective of a storage system, but it can
203
be tricky from the perspective of the server. Remember that additional

reconfiguration must be done in the server. If you extend a LUN on a
storage system, the host will not be able to see the added space until
you reconfigure the file system to spread over the added storage. In thin
provisioning, you can create a virtual volume without limitations in size.
If your application needs 100GB, in fat provisioning you would allocate
at least 175GB. In thin provisioning you can overallocate much more,
without paying for physical storage.
Thin provisioning is a functionality that provides virtual storage capacity.

It simplifies application storage provisioning by allowing administrators to
draw from a central virtual pool without immediately adding physical disks.
When an application requires more storage capacity, the storage system automatically allocates the necessary physical storage. This just-in-time method of provisioning decouples the provisioning of storage to an application
from the physical addition of capacity to the storage system. In addition to
simplified management, thin provisioning delivers better performance and
higher utilization of physically installed storage system capacity.
Without Dynamic Provisioning
With Dynamic Provisioning
Purchase of physical disk space capacity is based on

future requirements. The initial cost is high while a major
part of the disk space will not be used in the near future.
Determine the total capacity hosts will need in the

foreseeable future and purchase the minimum capacity
necessary. Save the cost of buying all capacity now.
Example: To provision five 2TB LUs to a host, 10TB of physical

disk space is required.
Example: When you store 200GB in each LU, you need to purchase
1TB only. Some margin needs to be taken into account.
Data size
200GB
LU
(2TB)
LU
(2TB)
LU
(2TB)
LU
(2TB)
LU
(2TB)
Data size
200GB
Virtual
LU
(2TB)
Virtual
LU
(2TB)
Virtual
LU
(2TB)
Virtual
LU
(2TB)
Virtual
LU
(2TB)
Physical capacity 1TB

Physical capacity
10TB
Figure 5.8 An example of how thin provisioning can help you save the cost of buying all the
capacity in advance. Thin provisioning is sometimes called dynamic provisioning.
204
To add physical capacity at a later time, you should analyze the actual
data consumption for each thin provisioning storage pool in the past.
With this data you can plan the schedule to add physical capacity to the
storage pool. Also, when the actual data consumption increases rapidly,
you can get notification through alerts sent from the storage system.
Thin provisioning limitations

We implement thin provisioning to avoid future administrative tasks
connected with the extension of LUN capacity and an installed file system. The extension of a file system can be especially tricky and time
consuming. With thin provisioning implementation, the file system will
be installed to cover the capacity of the virtual volume, which is overallocated and does not correspond with the installed capacity. When we
add capacity, we do not have to extend the file system. In thin provisioning implementation, the data path is as follows:
Application <-> file system <-> virtual volume <-> storage pool <->
parity groups <-> HDDs
When we format the storage area with a particular file system, this
file system writes metadata. Depending on the architecture and frequency of metadata writes, we can distinguish among file systems that
are suitable for application of thin provisioning and file systems that are
not suitable for thin provisioning. This is because the metadata itself
can occupy significant storage pool space. With implementation of file
systems such as UFS, JFS or HFS, the file system formatting occupies
the same capacity as the virtual volume. This effectively means that we
cannot overallocate the installed capacity. Suitable systems are, for example, New Technology File System (NTFS), VMware, XFS and VxFS.
Another problem that can occur in thin provisioning implementation is connected with data deletion. When we delete data stored on a
205
mapped virtual volume, this data should be removed from the storage
pool and the space it occupied should be freed for other virtual volumes. This does not happen. When you delete data, the space in the
storage pool is not automatically freed. The file system usually does not
delete data it just marks it as empty and allows overwriting. The storage system does not know this and therefore cannot free the storage
space in the storage pool.
The file system can be tuned for performance or for effective data
management. NTFS is performance oriented. That is why it does not remove deleted data completely; it just marks it as deleted instead. NTFS
also performs write operations where the actuator arm and I/O head
are at the moment. With frequent delete and write operations, more
and more storage pool capacity is taken and almost none is freed. This
is not very effective when using thin provisioning.
On the other hand, certain file systems prefer more effective usage of
storage space. For example, the VMware file system overwrites deleted
data, which does not consume additional capacity from storage pool.
The issues connected with deleted data and cleaned storage pool
are caused by the independent nature of the relationship between
thin provisioning functionality and the file system. When we are not
using thin provisioning, the file system operations do not cause trouble. However, customers often demand thin provisioning implementation, so the choice of file system needs careful consideration. To ensure
optimized utilization of the storage pool, storage system vendors implement zero page reclaim functionality. The storage system scans the
storage pool for used data blocks that contain only zeros. These blocks
are then erased and freed automatically. The problem is that zero page
reclaim is hardware functionality, which is again independent from the
file system. A file system usually does not overwrite deleted data blocks
with zeros. To make zero page reclaim functionality work, we have to
interconnect its logic with file system logic by choosing the suitable file
system and configuring it properly.
206
Because of its advantages, thin provisioning is very often implemented. To ensure good utilization of a storage pool that provides capacity to
virtual volumes, we need to choose a suitable file system. The file system
should support effective deletion of once used data blocks and it should
perform write operations by overwriting data blocks that were marked as
empty instead of using unused capacity. Storage systems offer zero page
reclaim functionality, which helps to solve issues connected with freeing
up deleted data blocks in a storage pool. The file system must support zero
page reclaim functionality.
Fat provisioning occurs on traditional storage arrays where large pools

of storage capacity are allocated to individual applications but remain unused (not written to) with storage utilization often as low as 50%
Thin provisioning is a mechanism that applies to large scale centralized
computer disk storage systems. Thin provisioning allows space to be easily
allocated to servers, on a just-enough and just-in-time basis.
Overallocation is a mechanism that allows server applications to allocate more storage capacity than has been physically reserved on the
storage array itself. This allows leeway in growth of application storage
volumes without the need to accurately predict which volumes will grow
and by how much. Physical storage capacity on the array is only dedicated
when data is actually written by the application not when the storage
volume is initially allocated.
Virtualization in storage system controller

Controller based virtualization is a functionality we have briefly discussed in previous chapters. Now its time to look at this technology in
detail. Controller based virtualization functionality is supported mainly
by robust enterprise level storage systems. This enterprise storage system allows configuring one or more of its front end Fibre Channel ports
as external. This Fibre Channel port then serves for connecting other
storage systems, midrange or enterprise, of various vendors. Manage-
207
ment and advanced configuration is then provided by the virtualizing

storage system. This can be especially useful when we have many storage systems from different vendors. Management of such infrastructure is very demanding and expensive because you have to pay for
more storage systems administrators, their training and their licenses
for advanced functionalities such as data replication. Controller based
virtualization unifies the management of heterogeneous storage system infrastructure and enables you to use functionalities available on
the virtualizing enterprise level storage system on all your virtualized
external storage systems. External storage systems provide their LUNs
to the virtualizing storage system. The virtualizing storage system then
sees these LUNs as if they were internal.
Windows
Host
Unix
Host
Windows
Host
Mainframe
SAN
Physical view of storage
Cache
Vol1
Hitachi Modular
Array
Vol1
Vol2
Vol2
Vol3
Hitachi Enterprise
Array
Vol3
Vol1
Vol2
Vol4
Vol5
EMC Array
Vol3
Windows
Host
Vol1
Vol2
Vol3
Vol4
Unix
Host
IBM Array
Vol1
Vol2
Windows
Host
HP Array
Vol3
Vol1
Vol2
Vol3
Vol4
Unused
Volumes
Mainframe
SAN
Virtualized view of storage
Cache
Vol1
Hitachi Modular
Array
Vol2
Vol3
Vol4
Vol1
Vol2
Vol3
Vol1
Vol2
Vol1
Vol1
Vol2
Vol3
Vol3
Vol4
Vol3
Hitachi Enterprise
Array
Vol5
Vol2
EMC Array
Vol1
Vol2
Vol3
Vol4
IBM Array
HP Array
Figure 5.9 Controller based virtualization of external storage, the physical and logical view.
Hitachi Virtual Storage Platform (VSP) is an example of an enterprise level storage system that
supports virtualization of external storage.
208
To illustrate how the virtualization of external storage system is

done, we will use the example of HDS Virtual Storage Platform, which
is an enterprise level storage system that supports controller based virtualization.
First you need to set one or more ports as external. An external port
is used for discovering volumes presented from the external storage
system. The external storage system cannot be connected to ports other than the external port. In order to set the port attribute to external,
you need to release the paths set to the port. The attribute of the port
where the paths are set cannot be changed to external. Before starting
the management operations, you need to know the ports whose attributes can be changed to external. Remember that we have already
learned that front end Fibre Channel ports can be set as initiator, target,
replication and external.
To use an external volume with Virtual Storage Platform, you first
prepare the volumes in the external system connected to the systems.
The volume in the external system should be between 64MB and 4TB.
If LUN security is enabled on the port, you need to add the WWN of the
external port on the VSP system to the host group with the volumes
that need to be set as external volumes.
The external storage system and Virtual Storage Platform external
ports are cabled to the SAN. You need to configure zoning to enable the
external ports to connect to the external storage system. Fibre Channel
Arbitrated Loop and direct connections are also supported. Externally,
storage can be attached directly via Fibre Channel or attached through
an Fibre Channel switch. Data can be copied or moved between internal
and external storage over Fibre Channel distances.
The external storage is then ready to be discovered by Virtual Storage Platform. The VSP system will see the LUNs on the external storage
system after the external storage discovery. Once the external storage
systems are discovered and the LUNs available, you can map volumes
in the external storage system as volumes in the local storage system.
209
Now the LUNs of an external storage are ready to be propagated to

hosts through the VSP systems. You only need to configure additional
parameters such as cache mode and inflow control.
Cache mode specifies whether the write data from the host to the
external storage system is propagated synchronously or asynchronously
(with or without waiting for successful write acknowledgement). All I/O
to and from the local storage system always uses cache. Write operation
are always backed up in duplex cache (Chapter 2, Cache Mirroring).
Inflow control specifies whether the write operation to the cache
memory is stopped or continued when the write operation to the external volume is impossible. If you enable inflow control, the write operation to cache is stopped and the I/O from the host is not accepted
when the write operation to the external volume is impossible. If you
select Disable, the I/O from the host during the retry operation is written to the cache memory even after the write operation to the external
volume is impossible. Once the write operation to the external volume
becomes normal, all the data in the cache memory is written to the external volume (all the data is destaged).
Improper configuration of cache mode and inflow control attributes
can negatively impact performance.
Virtualization of external storage enables you to view an entire SAN

as one storage pool, independent from its physical location. It masks the
differences between heterogeneous devices, and it allows you to use common storage management, data replication, data migration and lifecycle
management tools.
These benefits may sound intriguing, but remember that all the data
from external storage to the hosts go through one cache the cache that
belongs to the virtualizing enterprise system. A few external storages will
have negligible impact on cache performance, but there are certain limitations that need to be considered. Note that this functionality is often subject to licensing, which can be very expensive.
210
Virtualization of a storage system: logical partitioning

You have already been introduced to logical partitioning of blade
servers. Logical partitioning offers the possibility to divide one physical
piece of hardware into many logical resources. This functionality is also
available in some storage systems.
Logical partitioning in storage systems enables data center administrators to perform logical partitioning of cache, ports and disk capacity,
including external storage, on the storage system to create independently managed virtual storage machines. The private virtual storage
machines help to maintain quality of service by acting as dedicated resources that are independently managed and reserved for specific applications. The crucial capabilities of this functionality are:
Logical partitioning of a storage system restricts access to data
and resources from users and storage administrators to the elements of a single partition.
It also restricts access from users and administrators to data and
resources outside of their authorized partition.
It dedicates resources for exclusive use by specific applications to
maintain priority and QoS for business-critical application.
These capabilities result in advantages, such as data security improvement, application performance optimization, management and
utilization based on business requirements.
Logical partitioning is a storage system microcode based capability.
It is usually possible to dynamically reconfigure logical partition (virtual
storage machines) without disruption of service.
Logical partitioning can be particularly useful in large data centers
that provide services to companies that need storage. A virtual stor-
211
age machine can be configured to exactly match the customers storage

performance and capacity requirements.
When implemented within one organization, logical partitioning of a
storage system provides enhanced security (administrators can access
and configure only the assigned partition) and QoS management.
Sun
Microsystems
HP
IBM
Microsoft
Windows
FC/IP
SAN
4
2
1
Virtual Storage
Platform
Physical Storage Pool
High Perf.
99.99%
Thunder 9500V
100%
General Purpose
Lightning 9900V
Backup
Archive
Thunder 9500
SATA
Figure 5.10 Logical partitioning of a storage system. In this figure we see one storage
system (Hitachi Virtual Storage Platform) with three external storage systems that create
a virtualized storage pool. USP is then virtualized to provide two logical partitions private
virtual storage machines. Hosts are then able to access and use only the resources (cache,
ports and disks) assigned to the respective partition.
212
Automated tiered storage

At the beginning of this book we mentioned the data lifecycle and
how it relates to data tiering. By now you should know that frequently
used data is stored in high performance SAS or SSD drives, while archived data is stored on cheaper and less powerful SATA or tape drives.
Data tiers are groups of hard drives with the same performance characteristics. With tiering we can move data from a LUN that is created
of high performance drives to another LUN that consists of low performance drives. Some storage systems allow automation of this process.
You may ask, What does this have to do with virtualization? Automated tiering uses virtualized disk space a storage pool. This storage
pool is spread over several parity groups, each group containing hard
drives of one type we can have SSD parity groups, SAS parity groups
and SATA parity groups. Once we have a storage pool, we assign the
hosts with virtual volumes. Automated tiering is often implemented together with thin provisioning because of shared prerequisites (storage
pool). To remember the basics of data lifecycle management that represent motivation for automated tiering implementation, lets take a look
at what kind of data we have according to availability requirements:
Mission-critical data
This is the most important data; its loss will stop the business
from functioning.
Storage infrastructure for mission-critical data must provide high
performance, high availability and almost zero downtime.
Business-critical data
This data is still important, but its loss will not stop the business
from functioning.
Storage infrastructure must provide reasonable performance and
availability.
213
In the event of failure, the system must recover within 8hrs.

Accessible online data
This data is cost sensitive, in a large volume, with a low access
rate.
Storage system must provide online performance and high availability.
In the event of failure, the system must recover within 8hrs.
Nearline data
This type of data is cost sensitive and has a low access rate, and
large volumes.
The storage infrastructure must provide automated retrieval in
less than one hour.
Offline data
This data is usually archived data related to backup or compliance.
The storage infrastructure provides limited access and is cost sensitive.
It can take up to 72hrs to retrieve from the storage infrastructure.
Once created, most data is rarely or never accessed. It should be
moved off your most expensive storage tier to lower, less expensive
tiers. This is the premise and promise of data lifecycle management
methodologies and tools. But how do you determine what can be
moved and how to move it efficiently? Automated tiering functionality
solves this problem.
214
With automated (or dynamic) tiering, the complexities and overhead of implementing data lifecycle management and optimizing use
of tiered storage are solved. Dynamic tiering automatically moves data
on fine grained pages within virtual volumes to the most appropriate
media according to workload, maximizing service levels and minimizing
the total cost of storage system ownership.
For example, a database index that is frequently read and written
will migrate to high performance flash technology, while older data that
has not been touched for a while will move to slower, cheaper media.
No elaborate decision criteria are needed; data is automatically moved
according to simple rules. One, two or three tiers of storage can be
defined and used within a single virtual volume using any of the storage media types available for the particular storage system. Tier creation is automatic based on user configuration policies, including media
type and speed, RAID level and sustained I/O level requirements. Using ongoing embedded performance monitoring and periodic analysis,
the data is moved at the sub-LUN level to the most appropriate tier.
The most active data moves to the highest tier. During the process, the
system automatically maximizes the use of storage, keeping the higher
tiers fully utilized. Automated tiering functionality is usually available
only on enterprise level storage systems.
All the benefits of Dynamic
Provisioning
Dynamic Tiering
Futher simplified management

Dynamic
Provisioning
Further reduced OPEX

Better performance
Figure 5.11 Dynamic tiering (also automated tiering) uses dynamic provisioning technology
and thus inherits its advantages of simplified provisioning, capital savings and self-optimizing
performance maximization. OPEX is an abbreviation of operational expenses.
215
Automated tiering traditionally supports a maximum of three tiers

in a pool. The SSD-SAS-SATA hierarchical model is the most common.
The capacity of any tier can be added to the pool. Capacity can also be
removed from a pool. Automated tiering functionality fills top tiers as
much as possible while monitoring I/O references. The data is moved
from one tier to another in chunks called pages. In Virtual Storage Platform systems, one page has a fixed size of 42MB. Page placement is adjusted according to cyclical trailing of the 24hr heat map. I/O reference
monitoring essentially operates on a 24/7 basis. Following a cycle, we
begin a background operation to move or relocate pages to better fit
the I/O pattern. The average sustainable I/O rate of the tier is also monitored and calculated. The system avoids overdriving a tier. Sometimes a
tier may not be able to handle all the most heavily used pages, so it may
elect to keep some in a lower tier. This is unlikely to be a real problem in
most cases, but nevertheless, the system watches for this.
Pool
Data Heat Index

Tier 1
High
Activity
Set
Tier 2
Least
Referenced
Pages
Automated
Tiering
Volume
Normal
Working
Set
Tier 3
Quiet
Data Set
Figure 5.12 Dynamic tiering in Hitachi Virtual Storage Platform.
216
Automated tiering works on the storage level, without the participation of hosts. It does not recognize files; instead it divides raw data blocks
into chunks called pages. This is often called page level tiering.
20% of data accounts for 80% of the activity this is why automated
tiering increases utilization and optimizes costs. It moves active, frequently accessed data to high performance and expensive tiers, while storing
cold data in cheaper volumes.
Automated tiering is often implemented with thin provisioning. Both
technologies use virtual volumes that draw capacity from virtualized storage space storage pools.
Network based virtualization

Virtualization of servers can be implemented by using third party
software products or as inbuilt microcode functionality. Storage system
virtualization is again a functionality that is part of the microcode. To
virtualize external storage systems, we need an enterprise level storage system that supports this functionality. We know that this type of
virtualization has certain limitations. The major disadvantage is that we
have to purchase an expensive storage system capable of provisioning
virtualization services. In addition, we also need to purchase expensive
licenses to enable this functionality. Performance limitation is determined by the fact that data from all connected external storages flows
through one cache. If we connect a high number of external storage
systems, we can observe more or less significant performance reduction. To avoid this issue, we can consolidate our heterogeneous storage
system infrastructure by virtualizing it with a dedicated appliance.
A virtualization appliance can be connected in-band or out-of-band.
In-band virtualization is also called synchronous virtualization and outof-band virtualization is sometimes referred to as asynchronous. Well
217
look at the difference between the two shortly. First, lets look at the
similarities.
Both in-band and out-of-band virtualization controllers add intelligence to provide functionality available only in high-end enterprise level storage systems. These functionalities include mirroring, replication
and leveraging the connectivity of the SAN. Storage and SAN vendors
argue that intelligence and storage functionality belongs in the SAN (it
should be controlled by a dedicated appliance), and storage should be a
commodity. Both in-band and out-of-band solutions have no relevance
to the controller-based storage virtualization implemented in some
enterprise storage system. These solutions are thought to have an advantage over storage controllers since they map common functionality
across heterogeneous storage systems.
The general concept of in-band is that the mapping of virtual to
physical is done in an appliance that sits in the data path between the
host application and the target device. With out-of-band, the virtual to
physical mapping is done in an appliance that sits outside the data path
and the mapping is communicated over a control path to SAN switches
(or HBAs) that sit in the data path between the host application and the
target device.
In-band virtualization refers to its location in the storage network
path, between the application host servers and the storage systems.
An appliance or blade intercepts the I/O from the host, cracks open
the data packets to examine the original content directory blocks and
remaps them to another storage address. It provides both control and
data along the same connection path.
Out-of-band virtualization describes products where the controller
is located outside of the SAN data path. The virtualization appliance
communicates over a separate network with agents installed on the
hosts, intercepting and redirecting the I/O across the SAN. This type of
solution separates control and data on different connection paths.
218
In-band
virtualization engine
SAN interconnection
Data
Server
Out-of-band
virtualization engine
Storage pool
SAN interconnection
Data
Server
Storage pool
Figure 5.13 In-band and out-of band implementations of a virtualization appliance.
Cloud computing and storage

Virtualization establishes strong justification for transition to the
cloud model of service provisioning. In cloud computing, a large data
center offers a lease of infrastructure. In the data center, there is a large
number of servers (typically blades) and storage systems. Thanks to virtualization, these resources can be sliced into many virtual machines
that exactly match a customers requirements. The customer can rent
a number of virtual machines with clearly defined characteristics, such
as the number of processors, RAM, storage capacity and throughput.
This model of service provisioning is very flexible. The customer pays
only for the computing and storage capacity needed. Furthermore, the
customer does not need to worry about maintenance.
219
Private
Hybrid
Internal
Data center agility
Public
External cloud
Trusted cloud
External cloud
Regional Data
Center 1
Enterprise Data Center

Automation
Firewall and Security
Hybrids
Hybrids
Regional Data
Center 2
Public Cloud
Service Provider
Data Center
Private Cloud
Hybrids
Regional Data
Center 3
Regional Data
Center 4
Adoption
Figure 5.14 Cloud solution topology.
We differentiate among private, hybrid and public clouds. The private

cloud is located within a customers data center and provides greater
flexibility and faster time to value. Time to value (TTV) includes, for example, the effort, barriers, time, resources, planning and change that a
customer experiences before first seeing a tangible business value from
a product or service.
In a hybrid cloud, certain data or applications are extended outside a
customers firewall and moved to a trusted provider, stored either on a
shared infrastructure with multitenant capabilities or a dedicated infrastructure for each customer. The provider would have security practices
and auditable processes.
A public cloud is a publicly hosted facility that offers general access
to data. Generally, SLAs, security, and data availability and durability
cannot be guaranteed. Additionally, you may not have visibility or the
ability to audit your data once its inside the cloud. From an adoption
perspective (Figure 5.14), enterprise customers are adopting from left
to right even though a private cloud is more expensive than a public
cloud. As cost goes down, risk goes up, and vice versa.
220
The provisioning of cloud based solutions can be divided into three

distribution models:
Infrastructure as a Service (IAAS). In this model, the provider
is bound to provide the entire infrastructure. The entire infrastructure usually consists of servers (often in the form of virtual
machines), network and storage (virtual storage machines). The
main advantage is that the provider is responsible for uninterrupted operation and maintenance. The SLA between the provider and the customer defines the extent of the service, availability
rates and financial penalties the provider must pay in the event
of unexpected disruption or data loss.
Platform as a Service (PAAS). In this model, the provider takes
care of the development lifecycle and provisioning of web-based
applications and their modules. Software engineering is part of
this service.
Software as a Service (SAAS). In this model, the applications (software products) are licensed as a service that is leased to the customer. Customers are paying to access the application; they are
not paying for the application license itself. These applications
are hosted on the cloud and are available from anywhere there is
an Internet connection.
Cloud solutions provide major benefits to small and midsized companies, which can save a substantial amount of money that would otherwise be invested in the deployment of IT infrastructure. The provider
offers the whole infrastructure, its maintenance and configuration, including development of application and provisioning of software. The
disadvantage is that all the critical data is located outside the company,
posing potential security risks.
221
Virtualization: the cons

Having read this chapter you may have gotten the impression that
virtualization solves all problems and needs the world of IT is facing.
However, there are certain limitations and downsides that also need to
be taken into consideration before deciding to implement virtualization
of servers and storage systems into your organization. A major problem
can be the cost of software licenses. The more complex and powerful the solution is, the more it costs. For example, server virtualization
with blades is suitable for midsized organizations that need five or more
servers. A blade chassis alone can be very expensive, not to mention
individual blades.
Every virtual machine needs licenses for the operating system and all
additional software. If you purchase one physical server and virtualize
it, you need to purchase licenses for each virtual machine. You need to
think through your needs and calculate profitability of implementing
virtualization.
In storage system infrastructure virtualization, you can encounter
several problems. If you use an enterprise level storage system to virtualize external storage or if you use in-band virtualization appliance,
this type of solution can potentially become a performance bottleneck.
In both implementations of external storage virtualization, all the data
blocks from all externally connected storage systems go through one
cache and one controller. An out-of-band virtualization appliance can
prevent this bottleneck, but it requires a more complicated configuration. Virtualization of external storage can increase vulnerability of
infrastructure. If the virtualizing storage system or appliance fails, you
cannot use the data stored on external storage because only the virtualizer knows where the data is actually stored. Licenses that enable the
virtualization capabilities of the storage system are extremely expensive, and you need to calculate whether the virtualized implementation
is going to pay back.
222
Future of virtualization
Virtualization of IT infrastructure is becoming a mainstream trend.
We can encounter a certain degree of virtualization implementation in
almost every organization that uses an IT infrastructure. Its likely that
blade servers with native virtualization capabilities are going to spread
massively. The technology of Hyper-V Server is also undergoing major
development and can be expected to increase its share in the virtualization solutions market.
As for storage systems, we primarily use networked storage systems.
By networked storage systems we mean either a NAS system or a traditional block-level SAN storage system. Virtualization in storage systems
is frequently represented by thin provisioning and automated tiering.
In the future, its likely there will be an increased demand for storage
system logical partitioning as IT moves toward the cloud model of IT
services distribution. The capability to virtualize external storage is also
expected to undergo a boom as the need for consolidation increases. So
far, expensive licenses and storage systems capable of external storage
virtualization prevent large-scale implementation of this type of solution.
Market Adoption Cycles
Direct Attached
Storage
TODAY
Network
Storage
Cloud or Virtualized
Storage
Figure 5.15 Storage systems are heading towards fully virtualized solutions.
This chapter provided you with all the essential information on virtualization, its layers and areas of application. You were introduced to server
virtualization with VMware and Hyper-V software. You should know that
blade servers can be especially suitable for virtualization implementation.
Some blade servers offer advanced built-in virtualization capabilities we
call logical partitioning.
In storage systems, the prevailing virtualization capabilities are represented by thin provisioning and automated tiering. Both of these functionalities allow for greater flexibility, increased performance and improved
utilization of a storage system. They also provide simplified management
of your IT infrastructure.
High-end enterprise level storage systems can be used to virtualize external storage, which is very helpful when you need to consolidate your
heterogeneous IT infrastructure. Logical partitioning is extremly useful
when you want to ensure high performance and QoS of a critical application. Logical partitioning of a storage system is extremely useful in large
data centers that lease virtual storage machines for money. Cloud solutions can also use and benefit from this functionality.
223
224
Archiving and file and content

management

Fixed content and its characteristics
The difference between archiving and backup
What data objects are and how they differ from other data
How to describe essential components of a digital archive
Requirements of archiving solutions on storage systems
Features of content management
226
6 Archiving and file and content management
Introduction to archiving
In previous chapters, you learned that there are different types of
data. Traditional storage systems, both modular and enterprise, are
very efficient when processing structured data, or data in transaction based applications that work with large databases. This efficiency
is possible thanks to large, powerful cache and high data throughput
on front end and internal busses. Storage systems can handle a large
amount of I/O per second (IOPS).
Unstructured data includes text documents, worksheets, presentations, pictures or digitized material, such as scanned copies. This data
is represented by individual files. To store this kind of data, we can implement tiering that allows low cost storage of data that isnt accessed
frequently. It is also possible to implement a NAS device and share logical volumes of a storage system as network drives, making them easily
accessible to users. To effectively organize and manage your unstructured data, you can use content management platforms, such as Microsoft SharePoint or Hitachi Content Platform. A content management
application usually runs on a server and offers extensive possibilities
of storing, organizing and distributing unstructured data. We will talk
about content management platform capabilities in detail later in this
chapter.
Apart from structured and unstructured data, we have backup data
and data that needs to be archived. Backup data is traditionally stored
using low performance hard drives or tape libraries. The purpose of
backup data is clear in the event that we lose primary data, we can
restore it from a backup copy.
Archiving data is very different from backing up data. In archiving,
we always work with fixed content. Fixed content represents data that
will not be changed anymore. We archive data to meet legal requirements (as with medical records and accounting). We also archive data
because we want to protect older information that is not needed for
everyday operations but may occasionally need to be accessed. Data
227
archiving represents special requirements on storage devices. Archived

data needs to be retained for years or even decades. This data can be
further divided into online accessible data and offline accessible data.
Archive data can be either structured or unstructured. Structured
archive data is represented mainly by locked databases, which are consistent databases that will not be altered by adding or removing individual items. Unstructured data is represented by various files that are
not expected to be modified. It is only possible to modify archive data
by creating a modified copy. The original must be kept unchanged.
Fixed content
Fixed content consists of data such as digital images, email messages, video content, medical images and check images that do not
change over time. Fixed content differs from transaction based data
that changes frequently and is marked by relatively short usefulness.
Fixed content data must be kept for long periods of time, often to comply with retention periods and provisions specified by government regulations and law. Transaction data needs to be accessed very quickly.
However, access times for fixed content data are not so important.
Traditionally, fixed content data used to be stored on low cost tapes
or optical media usually with write once read many requirements,
while high performance storage systems were used mainly for transaction data. With the price of hardware going down and the application
of techniques such as automated tiering, nowadays its customary to
store fixed content data on low performance, inexpensive, online SATA
drives installed on storage systems.
Fixed content can be managed through a server running specialized
archiving software or by a dedicated appliance. In an organization with
many branches, an archiving appliance can be installed in the branch
office and feed the archiving solution located in the main data center.
228
This process is called data ingestion. Up to 50% of data that companies

use possess characteristics of fixed content. Implementation of fixed
content management and archiving solutions is therefore becoming a
standard, even for small and midsized organizations.
Legal Records
Satellite Images
Biotechnology
Email
Digital Video
Medical Records
Figure 6.1 Examples of fixed content.
Archiving can be defined as a task of maintaining data over its lifecycle.

It is typically used when we need to retain or preserve data for an extended period time. Compliant archiving is the ability to retain data securely
for a period of time dictated by a regulation. Archived data must be immutable (cannot be changed) and retrievable (in a timely manner). Archives
are designed to hold static data. Static data is also called fixed content.
Archived data does not require regular backup.
Other differences between archiving and backup

Generally speaking, backup generates a redundant copy of data.
The backup data can be deduplicated or compressed, or it can undergo
additional processing that ensures its compactness, but the original
data is retrievable through the procedure of recovery process. Archiving does not create redundant copies of data it only moves it to a
separate location designated for data archiving. Archiving makes data
backup easier. Only the portion of data that changes is backed up. We
are talking mostly about transaction data or files that are frequently
229
accessed and modified. Archived data does not need to be backed up

regularly. It can be secured by RAID protection, replicated or duplicated, but it is not subject to conventional backup techniques. Archiving
using a digital archive offers extensive features that ensure integrity
and authenticity of the data. In other words, it ensures that the original
file has not been altered. Remember that a backup is always a copy,
while archived data always contains original documents that cannot be
further modified.
Motivation for data archiving

There are several reasons for implementing data archiving policies
and technical solutions. Some of the most prominent reasons are:

Effective utilization of high performance tiers

Cheaper storage for fixed content
Data retention regulation
Simplified content management
Indexing and searching capabilities of a digital archive
Fixed content data should not be stored on high performance tiers

that use expensive SSD and SAS hard drives. By moving data from expensive storage media, an organization can save a significant amount
of money. A digital archive automatically recognizes fixed content data
and moves it to an archive location according to policies set by the administrator. Do not confuse this functionality with automated tiering.
Automated tiering can be used with live databases. Automated tiering
works on the block level and moves data among the tiers according to
frequency of access. All live data, meaning data that is often modified,
can be divided into parts that are accessed and modified frequently
and can be stored on low performance tiers. This is very different from
archiving fixed content. Fixed content is never modified and works on
the file level.
230
NAS Head
Storage
Tape Library
NAS
- No retention policies
- Limited protection from
user deletion of files
Archive
- No protection from early

deletion (WORM)
- Questionable retrieval long term
Optical Jukebox
RAID Array
- No protection from
user deletion (WORM)
- Limited grooming policies
(deduplication, search... )
- Limited WORM
flexibility
for deletion
- Older (expensive)
technology
Figure 6.2 Seeking an archiving solution. The storage systems we have discussed up until now
are not very suitable for fixed content storage and archiving.
Unlike traditional storage systems, digital archives are optimized for

long-time, low-cost data retention. IT technologies and digitization are
reaching almost all fields of human activity. Large quantities of data are
coming into existence every day. Digital data is now used in almost all
organizations and institutions, replacing documents in paper form, file
cabinets and traditional archives and depositaries. With the advancement of information technology, regulations and laws have emerged
to govern data management. Retention periods for particular kinds of
documents are getting longer. Digital archives need to meet these requirements. In Figure 6.3, you can see what kinds of documents are
legally bound to be retained and how long the retention periods are.
231
Retention time frames by industry

Life Science/Pharmaceutical
Processing food
Manufacturing drugs
2 years after commercial release

3 years after distribution
Manufacturing biologics
5 years after manufacturing of products
Healthcare HIPAA
All hospital records in original forms
Medical records for minors
5 year minimum for all records

From birth to 21 years
Full life patient care
Length of patients life + 2 years
Financial services 17a-4

Financial statements
3 years
Member reg. for broker/dealers
End-of-life enterprise
Trading account records
End of account + 6 years
OSHA
Sarbanes - Oxley
30 years from end of audit

Original correspondence 4 years after financial audit
Figure 6.3 Legal requirements of data retention periods. We can see that many organizations have increasing regulations, especially in the pharmaceutical industry, the food processing
industry, healthcare, financial services and auditing. The Sarbanes-Oxley Act very strictly regulates the length of retention of financial records and accounting in companies.
In most companies, the IT equipment for archiving is decentralized

and fragmented. Some portion of data is transferred to tapes; the rest
is stored in a NAS device or similar location. This fragmentation of an
archiving solution can make content management difficult because it
does not generally provide advanced search functionality across all archiving devices. In other words, it may be difficult to find the particular
document you want to access. A digital archive contains proprietary
software that allows an advanced search across all archived data.
232
Data creation
applications
Email Server
Document
Management
General Accounting
Web Application
Completely Duplicated/
Interconnected hot-site
Remote Disk Mirroring
More
Disk Mirroring
Shared Disk
Disk Consolidation
Single DIsk Copy
Electronic Vaulting
Tape On-site
Tape Backup
Off-site (trucks)
Search
#1
- Lack of scalability
Search
#2
SMTP
CIFS
Imortance
of Data
More
Search
#3
Amount
of Data
Less
Less
Delayed
Recovery
Time
NFS
Immediate
Search
#4
HTTP
- No search across
disparate storage
system
Storage
NAS Head
Tape Library
Optical Kukebox
NAS
RAID Array
Figure 6.4 An example of a decentralized and fragmented archiving solution. Disparate storage systems do not provide a common search engine, and they are not very scalable. A digital
archive can solve this problem.
Across various industries today, organizations are seeing the creation

of new fixed content (data) exceeding that of traditional data. Fixed content requires unique capabilities for effective storage, management and
preservation. Business organizations are being challenged to manage
fixed content more effectively based on rules and policies because of the
growing pressure to comply with regulatory mandates and/or information
governance for transparency. Difficulties in searching and retrieving content, as well as producing an auditable chain of evidence for authentic
records, is resulting in costly legal discoveries and audit processes. These
costs directly affect a customers bottom line in terms of economic and
operational costs (human effort). As a storage environment grows in complexity to handle the growth of fixed content, the cost of management
also becomes a significant challenge. Customers need a better method to
manage fixed content based on business policies and rules associated with
certain content types.
233
The right format of archived data

One of the first decisions we have to make is to choose the right
format for archived data. We can decide to store data in its native format. If we have common files with extensions such as .docx, .jpg, .tiff
and .pdf, we can store them as they are. However, we would also have
to ensure conservation of the applications able to read these files. As
application development progresses and new functions of text editors
and similar programs are added, the reverse compatibility with once
used standards can turn out to be a problem.
Another possibility is to use digital archive software advanced functionalities that transform our original file into a standardized file format. We can make such a setting that all Microsoft Word documents
and images marked as fixed content are automatically converted to
standard .pdf files. Another possibility is to convert the data into an
archiving data format developed either by a digital archive vendor or
third party software provider. The major problem with converting native file formats into standardized file formats is connected with authenticity of the original. This matter has not been fully resolved yet
and it is still unclear whether converted documents are still considered
as originals by law or whether the legal system sees them as altered
and modified copies.
Most companies prefer to archive data in its native format. A few exceptions can be named, though. Organizations that specialize in longterm data preservation can develop their own standard format. NASA
Data Active Archive Centers (DAACs) and the U.K.s National Digital Archive of Datasets are examples of organizations that always convert native formats into their own standardized file formats. The motivation
for implementation of a standardized data format can lie in the need to
attach extensive additional information about file origin. This information is called metadata. If the digital archive capabilities do not fit an
organizations needs, it can be the reason for implementing a standardized file format.
234
Digital archive: object oriented storage system

A digital archive is also called an object oriented storage system or
content aware storage system. Traditional RAID arrays are working on
the block level. The storage system itself does not know what information is stored in the data. It does not even work on the file level (except
in NAS devices). Data blocks (ones and zeros) are stored, striped across
hard drives or replicated without storage system awareness of what
kind of data the blocks represent. A digital archive, on the other hand,
is content aware. It is able to store data in the form of objects. As we
say that traditional storage systems work on a block level and NAS devices on a file level, digital archives work on an object level.
The object level is hierarchically above blocks and files. An object
consists of a file we need to archive. For example, it could be an email,
x-ray, MRI, photograph, satellite image or Word document. Aside from
the file itself, an object contains metadata. We know that file systems
such as NTSF also work with metadata.
The difference between file system based metadata and object
metadata is that a storage system administrator can define object
metadata. Metadata is encapsulated into the object and cannot be
separated from every individual file. In addition, an object also contains
definition of policies. These policies may include configuration of the
data protection level, indexing, retention periods, shredding and versioning. All these policies will be described in detail later in this chapter.
235
Fixed content data

Metadata
<M>
System information about the object

Customer-supplied metadata
Policies
Data protection level

Indexing
Retention
Shredding
Versioning
Blocks
Files
Email
X-rays
Photographs
Sattelite images
Microsoft Word documents
Microsoft PowerPoint
presentations
Object
<M>
Figure 6.5 A digital archive works on the object level. Each object contains fixed content data,
metadata and description of policies.
Hardware components of a digital archive are not very different

from traditional storage systems. In fact, a digital archive consists of
dedicated servers that run proprietary software supplied by a digital
archives vendor. These servers process the data and provide management console. Then the data is stored again on a traditional disk array
(modular or enterprise). The scheme is similar to a NAS operation. A
NAS solution also consists of a NAS head, which is a dedicated server
running proprietary software, and a block level storage system. Digital
archive servers are linked to storage system tiers equipped with low
cost and low performance hard drives. Aside from servers and storage systems, a digital archive may include other components, such as
data ingestion appliances that collect data from production servers and
send them to a digital archive over distance (through a LAN or WAN
connection).
236
Block Level Storage
Object Level Storage
<M>
LUN Based
Object Based
Primary Online Storage
Fixed Content Storage (long-term storage)
SAN Connected to Application
IP Network Connected to Application
High Speed
Object Aware
Huge Capacity
Policy Enforcement
LUN Level Access
Object Level Access
Figure 6.6 A traditional block level storage system compared to an object level storage system. The object-level storage system consists of powerful proprietary servers and management
software. These servers are connected to a RAID array.
A digital archive works with objects. An object contains actual data (a

file or files), metadata (data about files and their relationship) and configuration of policies that govern data archiving.
An object aware storage system consists of dedicated proprietary
servers and a disk array suitable for long-term data retention. Proprietary
software enables fast and efficient management of archive data. A digital
archive is fed data from production servers or through data ingesting devices.
Data objects and their components

To illustrate what a data object looks like, we can take a look at a
Hitachi Content Platform digital archive. In Figure 6.7, we can see a decomposed data object. Individual elements that have a particular function are represented by individual files, in this case .xml and .txt con-
237
figuration files. Xml represents a data format that was designed to be

readable by both humans and computers. The txt format represents a
universal text file. Both these file types ensures a high level of compatibility and standardization. The data to be archived is represented by
the My Document.doc file. Additional files contain policy configuration
and metadata. The file created.txt contains information about the date
and origin of the archived file. The file dpl.txt contains configuration
details for the data protection level. Other metadata files contain retention requirements, indexing information, the shredding policy and
hash data. Hashing serves for data authenticity validation. All files can
be set to comply with the WORM protection standard; they can be read
but cannot be modified or deleted. End users do not access objects in
the form depicted in Figure 6.7. End users typically use a web-based
interface that allows comfortable searching and previewing of archived
content.
fcfs_data
Familiar file system
interface
- Browse folders and
subfolders and
content
- Preview stored
content
- WORM protection
.lost+found
My Document.doc
fcfs_metadata
Original file name

and format
.directory-metadata
info
settings
My Document.doc
Metadata presented as
file system object
core-metadata.xml
created.txt
hash.txt
Authenticity established
with standard hash
algorithms
retention.txt
Retention period
managed for
each object
dpl.txt
index.txt
shred.txt
tpof.txt
Number of copies
maintained
Figure 6.7 A data object and its components in detail. This example illustrates how objects are
handled by a Hitachi Content Platform digital archive.
238
Digital archive features

A digital archive must ensure long time preservation and authenticity of stored data. High requirements of stored data authenticity are
met through active periodical verification of archived documents. Authenticity and credibility of data are key prerequisites, especially for auditing purposes. Additional functionalities of a digital archive increase
utilization and economical operation. Active functions of a digital archive are:
Content verification
Protection service
Compression service
Deduplication service
Replication service
Search capabilities
Content verification ensures authenticity and integrity of each data
object. A digital archive periodically checks whether the stored data
still matches its cryptographic hash value. When we store a document
in a digital archive, a special algorithm calculates its hash value. This
hash value is generated from the archived file and is unique for each
file. In simple terms, a hash value is similar to a finger print. The content verification has two main tasks: to detect corrupted data and to
repair corrupted data.
Data can be corrupted if the file is altered or modified accidentally
or purposefully, or if a digital archive component malfunctions. To detect data corruption, the digital archive periodically recalculates hash
values for all documents and matches them to the originally created
hash value. If the old and new hash values for a document are identical,
the document is not corrupted and its integrity is not endangered. A
violation occurs when the hash values do not match. In this event, the
239
digital archive must repair the data and return it to its original, uncorrupted state.
Corrupted data can be repaired from an existing good copy (such
as a RAID or digital archive replica). The digital archive creates a new
copy of the data from the good copy and marks the corrupted copy for
deletion.
Protection service is implemented in order to ensure stability of the
digital archive by maintaining a specified level of archived data redundancy. This level is called the data protection level (DPL) and is set by
the digital archive administrator. Different settings can be configured
for different types of archived content. DPL configuration tells the digital archive how many copies of the archived document it should keep.
Compression service can be implemented in order to achieve better
utilization of storage space assigned to the digital archive. As we described in Chapter 4, compression detects repeating sequences of ones
and zeros, searching for repeating patterns. These patterns are then replaced by pointers to the first occurrence of the pattern. Compression
effectiveness is determined by the level of data entropy. Please refer to
Chapter 4 for more detail.
Deduplication service, which is very well known from backup solutions, can be implemented in a digital archive as well. Unlike compression, which searches for strings of repeating bits, deduplication
searches for duplicities in files. Compression works on the level of ones
and zeros, while deduplication works on the file level. Its possible for a
single document to be marked for archiving from several locations. The
digital archive detects duplicities. For two or more files to be deemed
identical, their hash values must match. Deduplication can save significant amounts of storage space.
Replication service represents a recommended solution for ensuring redundancy of archived data. It is usually not possible to back up a
240
digital archive by conventional means because it can occupy terabytes

or even petabytes of storage. Traditional backup is also not well designed to create copies of fixed content. Replication service replicates
the data to another digital archive. This replica can be used for load
balancing. In other words, users can access a file on either system, according to its load. Replication service is linked to content verification
service. If the corrupted data is found, the replica can serve for data
repair.
Search engines allow users to search documents according to information stored in their metadata (such as who created the data, when
the data was created and the type of data) or using full text search
capabilities (searching for words and strings of words inside text documents).
Search is a faster way to find desired information within multiple storage resources. Indexing is a process that helps you search content much
more quickly. The indexing process creates an interpretation of the content and metadata of a file or object and stores that interpretation in an
optimized way for faster, more complete and accurate search results.
Search tools query the index for keywords, metadata and more. Results
are presented back through the search tool. A digital archive often provides indexing services that help search engines.
Digital archive accessibility

Digital archives should be easily accessible by common access protocols. These protocols include web-based protocols (http, https), network based file systems (NFS, CIFS), mail (SMTP) and file transfer protocols (FTP). Digital archives should also support application programming
interfaces (APIs) for command line interface (CLI) based access. In simple words, a digital archive should be accessible and manageable from
a command line. A command line interface enables scripting, which
241
leads to automation of management and operations. It also provides a

fast and flexible tool for simple installation and integration of a digital
archive. According to this information, digital archives should be accessible through:
Search console usually a web-based solution with graphical
user interface
Network as a file system usually by using an NFS or CIFS protocol; data is then displayed as depicted in Figure 6.7
FTP protocol an FTP client can connect to a data repository and
download documents directly
SMTP protocol useful for mail archiving
CLI command line interface
End users typically access a digital archive through a web-based GUI,
since it provides the simplest and most intuitive interface. End users
can also access the repository through an FTP protocol that is more efficient for accessing archived files.
Storage system administrators are likely to access the repository using NFS protocols or CLI.
NFS and CIFS protocols allow connection of a digital archive as a network drive on UNIX and Windows systems. SMTP protocol allows direct
connection between an email server and digital archive. Thanks to http
and https protocols, authorized users can access the archived data through
any compatible web browser.
242
NFS
CIFS
Compatibility
interface, primarily
for UNIX
Mount the HCP
file system path
(data or metadata)
High protocol
overhead
Compatibility
interface, primarily
for Microsoft
Windows
Map network drive
to HCP file system
path (data or
metadata)
SMTP
Archive email
directly from
email servers
HTTP(S) and WebDAV

WebDAV
HTTP(S)
Performance close
to HTTP protocol
(manage files of
Web servers)
Supports RFC 2518
compliant clients
Mount Points just
part of URL
Fastest protocol
Many good client
libraries
GET, PUT, HEAD,
operations
Can specify
metadata in URL
Figure 6.8 In this example, a digital archive can be accessed using four independent standard
protocols. WebDAV (Web-based Distributed Authoring and Versioning) is an extension to HTTP
protocol that allows remote management of files stored on web server.
Open standards based protocols provide applications and users a common and familiar means for interfacing to the repository. Each maintains
the familiar and ingrained file system metaphor people have embraced for
years and does not require changing the paradigm for how files and content are organized in storage. It also gives independent software vendors
(ISVs) options for integrating based on the best approach for their particular application and its specific performance and/or functional needs.
That means multiple applications, and potentially users, share a common
repository environment simultaneously.
Digital archive compliance features

Compliance can be defined as a state of being in accordance with established guidance, specifications, rules or legislation. A digital archive
needs to comply with such regulations. Digital technology is reaching
almost all institutions and organizations, and compliance requirements
are becoming stricter and more detailed. Compliance has increasingly become a corporate management concern. Devices such as digital
243
archives must offer functionalities that ensure a standardized level of

compliance and simplify compliance management.
So far in this book, we have already mentioned some techniques
that help to comply with regulations; we will discuss these now in
greater detail. There are also functionalities we have not yet discussed.
The most important compliance features of a digital archive are:
Write once read many (WORM)
Retention period definition
Data shredding
Data encryption
WORM functionality was originally introduced in tape libraries. The
tapes were designed in such a way that the stored data could not be
altered, modified, overwritten or deleted. This functionality is now also
available in digital archives. Each object stored in an object oriented
storage system can be amended with metadata information that enables WORM functionality. Once enabled, this functionality cannot be
disabled. WORM functionality is very popular in archiving solutions and
helps to comply with legal standards and regulations.
We have discussed retention periods several times in this book. You
have learned that various types of content can be subject to legal regulation with strictly defined retention periods, depending on the information stored in the object and the organization that created the data
(such as medical records and MRI images in a hospital). In a digital archive, each object can contain retention period settings. It is also possible to set policies that govern the way each file type is treated. For
example, it is possible to set a policy for all .tiff images to be retained
for ten years. Once the defined retention period passes, the document
is usually destroyed. Effective management of retention periods for different types of records is an important feature digital archives offer.
244
Data shredding is a data destruction utility designed to securely

erase a hard disk or digital storage device. When you delete a file on
your hard drive or USB pen drive, the data is not physically removed.
The data is just marked as deleted and is available for overwriting. This
means that specialized software or hardware can recover the data by
searching for and restoring blocks of data marked as deleted. To permanently and securely destroy data, it is therefore not sufficient to just
delete it. The shredding utility overwrites data marked for deletion
several times with randomly generated meaningless data, preventing
possible attempts for deleted data retrieval. Data shredding is a key
security utility that ensures deleted data cannot be restored.
Data encryption can be enabled in most storage systems and digital archives. Plain data is coded by a special algorithm. This algorithm
makes data unreadable to unauthorized users. Encrypted data need to
be decrypted before it can be accessed. Decryption of encrypted data
is possible using a password that triggers the encryption algorithm.
Storage systems implement this functionality to enhance security capabilities. Encryption is very demanding for the computing performance
of an installed CPU. Every block of data needs to be processed by the
CPU when we implement encryption, as well as when we need to access (decrypt) the encrypted data. This means that implementation of
encryption capabilities can have a negative influence on a storage systems performance. When you employ full-scale encryption, you can
expect a 10-20% performance loss, depending on the storage system.
Metadata in digital archives

Generally speaking, metadata is data about data. File systems use
metadata to store attributes for each file (such as read only or hidden file) and access permissions. Object oriented storage systems use
their own metadata that stores even more information, such as WORM
configuration, retention periods, owners, origin of the data, detailed
245
description. In an object oriented storage system, metadata is often

stored in the form of files. The format used for metadata storage is
very important. It should be an open standardized format with a clearly
defined structure. XML data format represents the ideal standard for
metadata storage. XML stands for Extensible Markup Language. This
markup language is defined by a set of rules for encoding documents
in a format that is both human readable and machine readable. Information stored in XML format is standardized, meaning it can be read
and accessed by third party software. Digital archives should provide a
standardized interface (API). XML contributes to extended compatibility with third party products and supports integration with such products.
Scalability, hardware designing and migrations

The amount of archived data is growing exponentially. A digital archive must allow simple extension of its capacity. This ability is commonly referred to as scalability. A digital archive must be easily scalable
in petabytes (PB) to ensure accommodation of large quantities of data.
Performance of digital archive must be ready to process such amounts
of data. Scalability should also allow you to reach full operability with
minimal capacity installed. Additional capacity is installed online without disruption of operation as more storage space is required.
A digital archive must be designed to retain data for decades. Certain types of data are legally required to be stored for indefinite period of time. It is obvious that a single piece of hardware cannot be
expected to function for decades or indefinitely. When we consider
rapid development of new technologies, we can expect that one device can provide uninterrupted operations only for a limited time a
few years or perhaps a decade before it becomes obsolete and its
maintenance too complicated and expensive. Due to these limitations,
each digital archive must support simple data migration by its design.
246
Digital archives are working with objects. The reason for this is to accommodate metadata and related information for each file or set of
files that constitutes an object. These objects are also ideal for future
data migration. Archived data is stored in the form of objects; the objects are then locked for compliance reasons. Individual objects are created in such a way that they are hardware platform independent. By
design, the objects must allow easy manipulation and migration to a
new hardware platform.
Integration of a digital archive

It is important to realize that a digital archive is just one component of the complete archiving solution. Similar to a NAS solution, a
digital archive consists of servers that run proprietary built-in software.
These servers are connected to a traditional block level storage system,
either modular or enterprise. The attached storage system is usually
equipped with low performance, inexpensive SATA drives. The digital
archive must be fed the data. This is called the application food chain.
A standalone digital archive provides cheaper, compliant and reliable
data storage. It also offers advanced search capabilities and policies
management. However, data we need to archive is managed by its parent applications (such as Microsoft Exchange or file servers). That is
why we need a component that interconnects the digital archive with
the parent application. This component goes through data stored by
the particular application and according to set criteria and it chooses
the candidate files for archiving.
This missing component, which is responsible for feeding the data
to digital archive, is called an archiving application. The complete archiving solution is then a chain with three components: application, archiving application and digital archive. An archiving application is linked
both to production servers and the digital archive. On the production
servers, there are dedicated agents designed to work with a particular
247
application. You learned earlier that archiving does not create copies
of data; it moves the original data to the archiving location. The data
taken from the parent application is moved to the digital archive and
replaced by a link reference to its location in the digital archive. Complete archiving solutions can be differentiated as file archiving, email
archiving, enterprise content management (ECM)/electronic resource
management (ERM) archiving, enterprise resource planning (ERP) archiving or health care archiving. Before choosing a particular model for
a digital archive, we need to think through the process of integration
into our existing infrastructure. The right digital archive must be capable of processing data from all important systems running on production servers.
Digital archive cloud features

Digital archive vendors often boast cloud support in their products.
In fact, this feature is represented by native support of virtualization.
Virtualization allows you to divide one physical digital archive into several logical digital archives. Many organizations, especially small and
midsized organizations not legally bound to retain large quantities of
data, do not have to purchase the physical digital archive. Instead they
can lease a virtual digital archive from a data center. The virtualization
capabilities in a digital archive allow the data center administrator to
create virtual machines with just enough performance to satisfy the
customers needs.
248
Figure 6.9 Virtualization of a digital archive. Digital archives are suitable for integration in
large data centers. Individual virtual digital archives can then be rented to small and midsized
organizations that wish to outsource digital archiving.
Policy based content management

Policy based content management refers to the automatically invoked behavior of a digital archive movement, retention and disposal according to defined rules. This functionality represents a solution for IT automation. By bringing automation to digital archiving, the
policy based content management helps ensure compliance with legal
and regulatory requirements. It also safely offers greater control and
responsibility to storage consumers.
Lets look at a few examples of how policy based content management simplifies routine administrators tasks.
Typical IT department:
Policies are defined in such a way that all files that have not been
accessed in 90 days are moved off Tier 1 storage. Any .mp3 files that
are not stored in specified directories are to be automatically deleted.
249
Healthcare:
All medical images are automatically replicated to meet regulatory
requirements. All treatment records are to be kept for 50 years after a
patient passes away. All billing records are to be kept for 10 years after
they are paid in full. All MRI images are automatically moved from Tier
1 storage to the archive after 30 days.
Financial Services:
Automation based on defined policies stores mortgage documents
on WORM media upon completion and store for the life of the mortgage plus 10 years. Content that has fulfilled all retention requirements
is deleted and purged.
Policies are defined using the management console of a digital archive. Each object contains information about related policies and their
configuration in the form of metadata.
Archiving and content management

So far we have discussed content management from the perspective
of storage systems. The term content management can, however, be
understood from different perspectives as well. In the context of storage systems, we are talking about specialized servers with proprietary
software connected to a traditional block level storage system, either
modular or enterprise. This type of solution is called an object oriented
intelligent storage system or a digital archive. It provides several types
of interfaces through which the content can be accessed. The users
typically access content stored on this type of solution through http
and https protocols using standard web browsers. The web interface includes an advanced search engine that, in cooperation with content indexing and information stored in metadata, delivers high performance.
This is content management, but it is necessary to remember that it is
always fixed content management. There are other solutions designed
250
to manage content that is not fixed content that is frequently accessed

and modified. This is called live data.
From the perspective of most companies, content management is
understood as a process of organizing files, sharing files and information within an organization, and managing workflow. It can be also understood as a traditional information system (IS). Specialized software
provided by third party storage vendors runs on production servers.
This software can use the internal hard drives of such a server or, in a
larger organization, a mapped LUN on traditional storage system connected in a SAN. NAS devices in a LAN environment can also be used as
a storage platform that supports this type of solution.
Remember that you need to distinguish between management of fixed

content and content management that focuses primarily on live data, or
data that is frequently accessed and modified.
In this book, we are focusing on fixed content management because it
is closely connected with storage systems.
The traditional view of content management can mean something
completely different and unrelated to the topic of storage systems.
Enterprise content management solutions

Before we close this chapter, you should have some insight into enterprise content management solutions of live data. We are not talking
about storage systems or digital archives now, but rather about software solutions. We will take Microsoft SharePoint as an example of an
enterprise content management system. Data in our organization can
be divided among several subgroups. We have live data represented
by databases, used by transaction based applications. We have fixed
content data, such as MRI images in healthcare or finished account-
251
ing reports. We also have live data that does not fit these categories.
Every organization that conducts business needs to work on project
documentation. Every organization also needs to share information resources among employees to ensure knowledge is available throughout the company. This documentation is handled very well by content
management solutions such as Microsoft SharePoint.
SharePoint is a software based solution independent of the storage system field. It is an application that runs on a production server.
Files that are fed into SharePoint are stored in the form of database.
This database can be stored on the internal hard drives of a server,
on a NAS device or on a mapped LUN in a traditional storage system.
The management console allows sharing of documents mostly text
documents, graphs, spreadsheets and similar files among groups of
users that are given permission to access them. SharePoint allows both
effective content management and workflow management. Project
documentation can be modified, reviewed and versioned by all people involved in the particular project. SharePoint is a file based sharing
platform that allows comprehensive and sophisticated organization of
documents in relation to users and individual projects or tasks.
Digital archives can support integration with Microsoft SharePoint.
Specialized agents are installed on a server that runs SharePoint. Once
the particular project is finalized and closed, the live content (project
documentation) effectively becomes fixed content, which is no longer
accessed frequently. An agent identifies such files and automatically
transfers them to the digital archive.
252
In this chapter, you were introduced to yet another type of storage

system. The portfolio of all the most common storage systems is now complete. You are now familiar with traditional block level storage systems, as
well as modular and enterprise systems. You know how NAS devices operate. Digital archives, or object oriented intelligent storage systems, are the
last product discussed in this book.
Remember that a digital archive consists of servers with proprietary
software or a traditional storage system. Data is stored in the form of objects that contains extensive metadata information. Objects are hardware
platform independent and can be easily migrated.
Basic functionalities of a digital archive include protection services,
compression services, replication services and an advanced search engine. Compliance features of a digital archive consist mostly of WORM capabilities, retention period definition, shredding services and encryption
services.
In the following chapter we will look at common tasks storage administrators have to face every day. This should provide you with a more practical perspective on storage system administration and data management.
Storage system administration

How to describe everyday storage administrator tasks
How to configure and monitor a storage system
What tools are available for a storage system administrator
Who is responsible for storage system implementation
254
7 Storage system administration
Who is a storage system administrator?

Unlike other chapters that have presented theoretical knowledge on
how things work in data management and storage systems, this chapter should provide you with more practical insight on how things work
in real situations. To be more specific, this chapter focuses on people
who install and configure storage systems, as well as people who conduct maintenance tasks. This focus gives you yet another perspective
on storage systems management.
Before we start with the description of a storage system administrators duties, it is important to realize who is considered a storage system
administrator. A storage system administrator is an employee in charge
of maintaining a storage system infrastructure. Vendor policies determine the extent of this employees duties a storage system administrator has a specific set of tasks based on the service level agreement
(SLA). To make sense of this, you need to understand that there are two
kinds of people who perform administrative tasks on a storage system.
First, there is always a customer engineer from the vendor this is
the company that supplies the devices. This company also provides a
storage system implementer, who assembles, installs and integrates
the purchased devices. The implementer is specially trained to do these
tasks. The service contract strictly lists the number of operations a storage system administrator is allowed to perform without breaking service contract maintenance conditions. Under a warranty agreement,
the storage system administrators can perform only a limited number
of operations on the particular system. The more complicated tasks are
performed by a storage system implementer, who works for the company that supplied the equipment and has a partnership contract with
the vendor. A storage system administrator is an internal employee of
the company that purchased the devices from the vendor through the
vendors partner.
As we mentioned earlier, vendor SLAs determine the extent of a storage admins duties. Each vendor has different policies. Some vendors
prefer that the storage implementer does almost all the tasks, leaving
255
a storage system administrator with very few duties. (In this situation,
whenever you need to change the configuration of your storage system,
you have to call the support line.) In contrast, some vendors prefer for a
storage system administrator to do all the tasks while the implementer
performs only the equipment installation. As a result, we will discuss
only the most common storage administrator tasks. At the end of the
chapter, we will also discuss the most common storage system implementer tasks.
Storage administrator tasks

To illustrate the duties of a storage administrator in this section, we
will use the example of a growing organization. Lets say the organization has an existing IT environment but has not yet required an advanced storage system. Now the amount of data that needs to be handled increases to a state not sustainable without an advanced storage
system delivering high performance, availability and reliability.
As the project moves forward, the organization buys new servers
and storage systems. The internal administrator for this organization is
a specialist mainly in servers. (A LAN environment with servers is very
common, unlike a SAN environment, which requires additional specialization.) Because the administrator is aware of the importance of data
and its secure storage, he recommends purchasing a midrange storage
system for a good price and value ratio. The administrator makes a list
of storage system requirements and helps find a suitable vendor. He
must carefully consider the performance, availability, configuration and
required advanced functionalities. Once the administrator finds a suitable device and concludes the purchase, he must specify the configuration requirements and overtake the administration of the storage system from the supplier.
256
The configuration requirements must include details on how much

capacity is required, what hard drives are to be implemented and their
performance level. The administrator must assess the importance of
various kinds of data, determine what kinds of data need to be stored
on high performance hard drives and what data needs to be replicated.
The level of protection must be decided upon especially what RAID
level is to be implemented. The supplier provides consulting services
that should simplify the procedure for the storage system administrator,
who does not yet have extensive knowledge on storage system operations and SAN environments. The consultant from the supplying company gathers details needed to integrate a storage system from the administrator. These details include specifications of the host bus adaptor
(HBA) used in servers, the HBA firmware version, the operating systems
running on the servers and more. The supplier also provides consulting
on operation disruption prevention the required redundancy level of
individual components and the location for replicated data.
The storage administrator also participates in the preparation of additional documentation. The essential documentation and planning includes details on:
Capacity management
Availability management
Continuity management
Financial management
In storage system management, capacity can be considered a basic
resource. Capacity management deals with the amount of data that
needs to be stored, the size and performance of LUN per application
(server), the performance of individual hard drives, the performance of
I/O and R/W operations, and content characteristics.
In availability management, the administrator would configure and
sustain protection against component failure. Failure may occur in a
hard drive, controller, on HBA, Fibre Channel switches and other com-
257
ponents. Availability management also includes replication and backup

strategies, as well as data archiving. Data archiving should prevent data
loss due to inconsistency caused by application failure.
Continuity management is often part of business continuity planning and disaster recovery procedures. The administrator would need
to decide what threats he wants to protect his organizations data
against (such as fire, theft or floods). Existing business continuity and
disaster recovery documentation needs to be updated every time the
IT environment changes.
The administrator also needs to calculate the financial pros and cons
of new technology implementation. Financial management would include budget preparation, cost calculation and invoicing. Total Cost of
Ownership (TCO) should reflect all costs connected not only with purchase of the equipment, but also with its operations and maintenance.
TCO is a very important value that helps determine the real value of an
integrated solution.
Once all the documentation is ready and the supplier and purchasing organization agree upon all terms and conditions, installation of the
storage systems follows. In our example, and in most cases, this process
is completed solely by the supplier. The supplier also provides system
management training for the administrator, covering basic system operations, system monitoring and performance tuning.
When the device is integrated and operational, the supplier provides
services such as maintenance or monitoring according to the SLA. The
supplier then monitors the devices operations remotely. If the supplier detects a problem, the supplier contacts the administrator and discusses the solution. The storage administrator has undergone the usual
training to perform basic LUN operations (discussed in the following
section), monitoring and performance tuning.
258
Storage system administration in a SAN environment is different from

the common tasks IT administrators face. That is why a storage system
administrator needs to be trained to specialize in the field of storage systems. Storage system administrators are required to formulate demands
on the new equipment. They must choose the provider and specify the requirements on performance, capacity and additional functionalities. Once
the storage system is installed and operational, supplier policies determine the extent of operations a storage system administrator is responsible for. Storage system administrators are usually allowed to perform basic
operations with LUNs. Storage system administrators are also responsible for monitoring, basic performance tuning and infrastructure growth
planning In addition, storage system administrators are responsible for
maintenance and monitoring of SAN networks, network performance
and network security. They take care of the configuration of Fibre Channel switches, ports and zoning.
Common operations with LUNs and cache

In addition to SAN maintenance and data management, the storage
administrator is responsible for storage system monitoring and communication with the supplier, as well as basic operations with LUNs and
other configuration tasks according to the SLA and the administrators
training level. Basic LUN operations and configuration tasks include:
Configuration of RAID groups and volumes (Figure 7.1)
Implementation of changes in volume configuration
Data replication optimization
Configuration of cache
Cache partitioning
Backup of storage system configuration
In this sections example, lets say you are the storage administrator
performing common operations with LUNs and cache. To configure a
259
RAID group, you need to choose physical hard drives that will be linked
together in a RAID array, and you need to select the desired RAID level.
Depending on your requirements, you will probably choose a RAID-5
level with a distributed parity that protects against one hard drive failure. If higher data protection is desired, you may choose a RAID-6 level,
which protects against two hard drive failures. If the highest possible
performance is required, you may even choose RAID-1+0, which implements data striping and disk mirroring. There is no parity in this RAID
level. Therefore, there is no need for data reconstruction if one disk
fails, since you always have a fully operational mirror disk. To balance
performance and cost, you would probably go with RAID-5 or RAID-6.
1. Click RAID Groups
3. Select the desired configuration

and then click OK
2. Click Create RAID Group
Figure 7.1 Configuration of RAID groups and

LUNs. The picture depicts
the interface of a storage
system by HDS. The application (Hitachi Storage
Navigator 2) with GUI is
installed on a server connected to the storage
system.
260
Once you have created RAID groups, you need to decide your LUN
allocation policy. You have to decide especially whether you prefer fat
provisioning or thin provisioning. In the first case, you create a LUN and
map it to the server. For thin provisioning, you need to set up a storage
pool, create virtual volumes and map the volumes to servers.
As a storage administrator, you need to consider which front end
ports are used for mapping LUNs. To balance the load, the number of
LUNs linked to servers should be even for each port. If you have eight
LUNs and four ports, you should configure LUN mapping so that there
are two LUNs per port. This important step allows you to achieve high
performance, and it prevents bottlenecks that might have occurred if
you had linked all eight LUNs through only one or two ports.
In the case of fat provisioning, you should know how to enlarge the
existing LUN. There are various tools to do so. First the changes in configuration must be made in the storage systems management console.
The existing volume is enhanced. To complete this task, you also need
to enhance the file system to spread over the volume.
Imagine you have created a 10GB volume and mapped it to a server,
which formatted the volume with a file system. When you enlarge the
LUN on the storage systems level, the LUN will have, lets say, 20GB, but
the file system is still capable of accessing only the originally allocated
10GB. To overcome this issue, you need to perform file system extension operation. At the end of the procedure, you have a 20GB LUN formatted with a file system that is able to use all 20GB. In thin provisioning, when you get an alert that signifies the virtual volume and storage
pool are running out of capacity, you need to contact your supplier for
installation of additional physical hard drives.
Storage administrator tasks connected with data replication are
often limited. Usually, the supplier creates and configures replication
pairs, and you can make adjustments, such as changing the frequency
in which snapshots are created. The default supplier setting (according
to organization requirements) is that snapshots are created every 4hrs.
261
If this becomes insufficient, you can change this to 2hrs. You can also
perform pairsplit operations with in-system replication to be able to access data replica located on an S-VOL for the purposes of data mining or
data warehousing. However, you cannot usually create new pairs or set
up and configure remote replication pairs.
Cache optimization is one of the most difficult tasks you will face as
a storage administrator. Cache greatly influences the performance of
the whole storage system. We know that hard drives, even with implemented data striping, are still the slowest component of the storage
system. Cache is designed to compensate for the performance differential between hard drives and solid state components. Proper configuration is, therefore, essential. You can adjust certain parameters of cache
to enhance I/O performance, which is the key for transaction based applications that use large databases. According to the database type, you
need to choose the right block size the cache is handling. If a database
uses large chunks of data (512K), the cache setting must correspond.
When a database works with small chunks of data (16K), you need to
adjust the cache settings as well to achieve the best performance and
utilization. It is also possible to prioritize certain hosts. I/O requests
coming from such hosts are processed with higher priority. Additionally,
you can divide cache into logical units when you virtualize the whole
storage system (one storage system is divided into several logical storage systems), or when you need to support volumes under high load
and the character of the load requires large cache. It is, therefore, possible to assign a certain portion of cache to a particular volume.
You can export the configuration of the whole storage system into
a single file. You should export regularly and back up the configuration
files. If the configuration is accidentally altered, it is your task as a storage administrator to return the configuration to its working state, which
is easy to do using the backed up configuration files.
262
Monitoring of storage systems

Once the storage system is fully installed and configured, its operations do not require the high attention of a storage system administrator. Everything is fully automatic. Furthermore, all the components
in a storage system, as well as all the components in a SAN, are fully
redundant, meaning the failure of one component does not affect the
operation of the whole device, nor does it cause downtime. A working
storage system is not reconfigured every day; once it is fully integrated,
it can work for months without requiring the attention of the administrator. However, the administrator should monitor storage system operation regularly. In a storage system, the administrator monitors two
major things:
Performance of a storage system
Health status of all components installed in a storage system
Performance is monitored by measuring the IOPS. The data flow can
be measured on HBA ports in a server, on ports in the Fibre Channel
switch or on the front end ports of a storage system. Other parameters
that are monitored include utilization of cache, the CPU load and the
load on ports. Monitoring is very important because it helps detect possible bottlenecks. Log files with measurement values also represent a
necessary source of information required for performance tuning. (Performance tuning is discussed in detail in Chapter 8.)
263
Array group 4-5

approx 78%
utilization
Figure 7.2 Among IOPS on ports and CPU load, the storage administrator also monitors effective utilization of installed capacity. Here you see an example of a RAID group load and performance monitoring.
The performance of a storage system is measured through various

tools. These tools are usually supplied by the storage system vendor.
Some software requires the administrator to interpret values, and some
software provides automated monitoring. Once a bottleneck is detected, the software sends an alert message to the administrator, who evaluates the situation and ensures the problem is solved. Monitoring can
also be provided by the supplier, depending on the SLA.
The health status of all components installed in a storage system is
also monitored. A storage system is programmed to detect any possible
264
hardware failure before such a failure happens. Even individual hard

drives include these monitoring functions, which are built into firmware
(Self-Monitoring, Analysis and Reporting Technology or SMART). A storage system can read and evaluate values provided by hard drives and
detect a decrease in performance of a particular hard drive. Once a possible failure is detected, the system sends the storage administrator an
alert message. Other components are monitored as well. An alert is also
sent when a LUN is near capacity and needs to be enlarged. The storage
system administrators responsibility is to log the alerts and warnings
and deal with them appropriately. The storage system administrator
never takes care of the replacement of individual failed components.
The administrator also cannot replace a broken hard drive with a new
one or add new hard drives or expansion units. This is always the suppliers responsibility.
The last thing that requires the regular attention of a storage system
administrator is keeping all firmware and software up to date. Whenever there is a new version of firmware installed in a storage system,
the administrator should receive this firmware from the supplier. Either
the supplier (implementer) or a storage system administrator then performs the firmware update, which can be disruptive (requiring downtime) or nondisruptive (with no downtime). Modern high-end storage
systems usually allow nondisruptive firmware updates.
Note that firmware stored in individual hard drives may need regular
updates. Management software with a GUI and CLI also needs to be
kept up to date.
265
Checking
Operating
Action
Alarm
Generates email, SNMP

trap, or executes script
Setting
Create alarm definition
To set alarm definition, Performance Reporter
provides Alarms Settings GUI
including import/export alarm definition capability.
Figure 7.3 Automated monitoring functionality. You can define situations that trigger an action once they occur. Whenever performance and hardware health values reach the defined
status, an alarm (alert, warning) is sent to the storage system administrator. The storage system
administrator assesses the situation and ensures the problem is eliminated.
Storage system management software

The management software of a storage system is stored in microcode. However, the microcode itself does not provide any advanced GUI
or CLI. To manage disk arrays, you need to install a management software client on the server connected to the storage system. The management software then communicates directly with the storage system
microcode. The storage system implementer and administrator make
all the necessary configurations through this client. Usually, the configuration is done through the GUI, while the CLI is used mainly for scripting
that allows task automation. Scripts that use a CLI are especially useful
in replication. By executing the script, you can bring the database stored
in the cache and LUN to a consistent state and then perform a pair-split
operation. Scripting allows you to do much more than these tasks, but
it can be difficult to master.
266
The storage system administrator helps to design and maintain the

SAN infrastructure. For example, the administrator helps to specify the
requirements on storage systems and choose the right product and vendor. The day to day tasks mostly include monitoring and slight configuration changes. However, the amount of data that needs to be processed
is growing rapidly. To accommodate such growth, the storage system
administrator needs to plan expansion of the existing infrastructure, including the purchase of new storage systems, Fibre Channel switches
and servers. This time, the administrator can decide to purchase a disk
array from a different vendor because it fits the organizations requirements better than the vendor who supplied the first storage system.
The administrators major concerns are issues connected with the management of two different storage systems. With advanced software and
technology, these issues can be overcome. The management of a heterogeneous storage system administration can be implemented. This
can be done either by virtualizing an external storage system or by network and server based management (an appliance sits on the network
between the servers and storage systems). Independent management
software, either server based or network based, can:
Centralize administration and operation for a heterogeneous
storage environment under a common management application
Automate discovery and visualization of a storage infrastructure
and network topology
Analyze all path interdependencies between host servers and
storage systems
Track heterogeneous storage capacity usage so available capacity
can be moved where it is needed most
Integrate with leading enterprise applications to accurately measure storage usage by application when used with host based
agents
Offer modular design for easy integration with advanced storage
functions, such as chargeback and storage resource management
of file systems
267
Benefits of such management software include:

Save both costs and IT resources required to manage multivendor storage networks
Simplify heterogeneous storage administration by utilizing a common standard based management platform
Ensure storage availability for mission critical business applications
Identify underutilized capacity to defer additional storage investments
Accelerate the deployment of advanced storage functionality
without adding administration complexity
Figure 7.4 Command line interface. An example of a Hitachi Dynamic Link Manager configuration. This software takes care of path management.
268
Integration of a new storage system

When purchasing new devices to be integrated into a SAN or LAN
environment, the storage administrator must think through the whole
implementation process, including the following items:
The administrator must decide what kind of storage system and
connection the solution requires. A traditional midrange or enterprise storage system would be integrated into the SAN.
The administrator must make sure the existing Fibre Channel
switches have enough ports to service the new device. If not, the
Fibre Channel switches need to be upgraded.
Since the whole technology uses optical wires, the cabling needs
to be considered as well.
Switches must be equipped with small form factor pluggable
(SFP) transceivers.
The vendor may supply a rack and rail kit for equipment installation, but this option might be more expensive than using an
existing rack.
The electricity infrastructure must meet the new devices power
requirements.
The server rooms air conditioning may also require an upgrade.
If a NAS device or a digital archive matches requirements on the
new equipment better than a traditional storage system, the administrator must consider implementation of such a device within a LAN infrastructure.
Integration of a new device does not always have to be tricky and
complicated, but it requires careful planning and design. The cost of
training and maintenance need to be considered as well. TCO is a major
factor in deciding which vendor to choose.
269

Effective allocation of resources and reporting is a major concern
when you wish to provide a storage system as a service. Nowadays,
many companies run large data centers focused primarily on data storage. These data centers are equipped with dozens or even hundreds
of storage systems and digital archives. When you provide storage as
a service, the effective allocation of resources is very important. Virtualization enables the administrator to create made-to-measure virtual
storage systems that match customer requirements exactly. SLA contracts with the customer are usually very strict in terms of availability
and prevention of data loss. Therefore, effective allocation of resources
and their utilization can significantly affect profitability of the business.
Storage system management software allows exporting of various
reports that can be very helpful when preparing SLA contracts. These
reports are also an inevitable part of documentation required for auditing and compliance with ISO standards. The storage system administrator should be aware of this requirement and collect and manage
reports regularly.
Security management
Part of the storage administrators job is to make sure the data is
stored securely. A SAN environment provides inherited security features, mainly because of its isolation from LANs requiring more complex
security management. LUN mapping itself can be seen as a security feature because it links a LUN to a designated HBA (according to its WWN)
through a selected front end port. Other servers cannot see this LUN.
Zoning implemented on Fibre Channel switches is another common
method of security enhancement. Zoning is described in detail in Chapter 3. It is also possible to use data encryption on the disk array, but this
can affect its performance. In the case of NAS devices connected directly
270
to a LAN, it is desirable to implement a virtual LAN (VLAN) policy. A LAN

network can be partitioned into many VLANs. Devices from different
VLANs cannot communicate with each other. Authentication services
may be provided on a LAN network. An Active Directory Lightweight
Directory Access Protocol (AD/LDAP) server is a typical example of this
solution. This server verifies the identity of a user who attempts to access data stored on a secured device. In a LAN, other components can
be implemented to prevent cyber-attacks from outside well known
hardware firewalls have this feature as a standard. To keep both LAN
and SAN environments safe and secure, it is important to update all
software and firmware regularly.
Tasks of a storage system implementer

The duties of a storage system implementer can be very different
from those of a storage administrator. A storage administrator works
in the organization that buys equipment, while the implementer works
in an organization that sells equipment. Storage system administrators
have limited access to information, while implementers have all the
technical documentation at their disposal. This documentation includes
maintenance manuals and best practice guidelines. The implementer is
also allowed to attend partner trainings (because the company is a partner of a vendor). When a demand from a customer arises, the implementer in cooperation with the IT architect, IT consultant and pre-sales
team helps to design the best solution for the customer. The implementer is certified and can perform the installation and initial configuration of a storage system. The storage system administrator cannot.
The implementer has a deep knowledge of SAN environments and storage systems and, as a result, is more knowledge than a storage system
administrator.
The implementer performs the installation of the equipment and the
initial setup. The implementer also provides basic training to familiarize
271
the customer with the new device. The storage implementer is authorized to conduct all hardware replacement and upgrade procedures. If
a disk or other component fails, it is the implementer who can replace
the malfunctioning part. The implementer also monitors the storage
system remotely and, in cooperation with a storage system administrator, helps with performance tuning and advanced operations. Microcode updates are also the responsibility of a storage implementer.
This chapter provided you with a more practical insight into the duties
and responsibilities of a storage system administrator. Remember that the
storage system administrator is responsible for the whole SAN environment and should be able to configure Fibre Channel switches and server
HBAs, as well as perform basic storage system operations. By basic operations we particularly mean the creation and configuration of volumes
and their management. A storage system administrator also monitors the
operation of a storage system to detect possible malfunctions or performance bottlenecks. The storage system administrator is the person responsible for data management, design and infrastructure upgrades that
ensure data storage.
272
Performance tuning,
optimization and storage
consolidation

How to determine whether a storage system is properly
tuned for performance
How to increase and optimize the performance of a storage
system
What tuning possibilities we have for individual components
of a storage system
What values are measured and how to interpret them
Motivation and goals for consolidation
274
8 Performance tuning, optimization and storage consolidation
Motivation for performance tuning

When we purchase and implement a new storage system, the supplier should suggest a configuration that matches the requirements
specified in the project documentation. From this perspective, everything works and there should be few items that require immediate
attention. However, the amount of data our organization handles can
grow quickly. The requirements on our IT infrastructure may change
too. The characteristics of load can change in ways we might not anticipate. For example, a server previously under an average load of
30% must now handle more requests. When such situations occur, it is
necessary to adjust our IT equipment to these new conditions. Regular
monitoring and measurements should provide us with a large amount
of input data that needs to be interpreted correctly and transformed
into new specifications on the configuration of our storage system, SAN
or servers.
We can try to adjust the configuration to match the new requirements. However, if we make the adjustments improperly, we may encounter many significant problems. Once we change the configuration,
some LUNs may show degraded performance. The whole storage system may sporadically freeze for a second or two. Under high load, an
application requires extreme storage system performance. Storage systems may not be able to handle such peak situations effectively. The
application may then show problems that impact the end user.
All these issues can arise when we try to tune the storage system
and fail or when our IT environment changes and we do nothing at all.
Proper monitoring, performance management and storage system tuning are, therefore, vitally important. The storage system administrator,
in cooperation with the supplier, needs to be able to deal with problems that stem from changes in the IT environment.
275
Introduction to performance tuning

The storage system administrator is responsible for proactive management, which includes performing regular monitoring of the systems
and their components. There are several questions that can be answered with careful monitoring of resources:
Which systems are underutilized?
How much excess capacity is there?
Which systems are overutilized or overallocated?
What bottlenecks exist, and are they easily resolvable?
Can I be automatically notified when warning thresholds are
reached?
The major indicator of storage system performance is I/O response
time. If we get a good response time for all I/O coming from all servers,
then everything is working smoothly. When the I/O response time is
longer than expected, we need to search for the source of the problem.
Before we get to that, lets take a look at a typical online transaction.
The I/O response time equals the time that is necessary for processing
one online transaction.
30ms (.03 sec)1
CPU
Time
Queue
Time
I/O
Time
25%
15%
60%
This does not include network time (LAN or WAN)
Figure 8.1 Typical online transaction processing. A single transaction requires a certain portion of CPU time. Then it queues and is processed further. At the end of the whole process, the
transaction is concluded. In other words, the requested data or acknowledgment of a successful
write operation is sent back to the host. Notice that the average time for processing a single
transaction is 30ms. This does not include the time necessary for delivering data to its destination over the network.
276
When the I/O response time takes longer than it should, the storage administrator must determine where there is a problem. The simplest performance tuning is based on an even distribution of load over
storage systems and their components. If one controller is overloaded,
then the first option is to reassign the problematic volume to a different controller or to enable load balancing. The same counts for front
end port traffic. When one port becomes overutilized, redirect the data
traffic over a different port.
Storage system vendors offer software tools for performance tuning. These tools are designed to cooperate with the most often implemented applications and databases that run on the production servers.
These tuning managers can analyze load characteristics and suggest
changes in configuration. Modern storage systems allow nondisruptive
performance tuning. When we change the storage system configuration, it usually does not happen quickly. For example, when we change
the RAID level type, enable load balancing or change LUN characteristics, it takes time to implement these changes (especially because all
stored data needs to be moved to a new location). Nondisruptive performance tuning is, therefore, a very important feature.
However helpful the tuning managers are, their area of application
is usually limited to storage systems. The bottlenecks may occur on a
SAN as well. The tuning managers provide only monitoring and are able
to recommend changes in configuration. But in the end, it is always
the storage system administrator who must decide which changes to
implement and how. To complete this task, the storage system administrator must have the tools to measure the most important characteristics and the knowledge to use them. The administrator (or the storage
system implementer, depending on vendor policies) must have tools
for implementing a new, improved storage system configuration. The
administrator must also be able to write scripts that can provide automation of the most common performance tuning tasks.
277
Monitoring, analysis of operations and performance tuning are all part

of performance management. In performance management, we try to ensure efficient and optimized operation of our IT infrastructure. Monitoring
provides us with important values and data. Monitoring is a continuous
activity. For performance tuning, current performance values are of little
use, as we need to collect performance data over a longer period of time
weeks and months. Performance analysis evaluates this data and provides us with clear information on whether all the components involved in
data processing meet expected performance levels.
Performance tuning then implements data collected during the monitoring phase and data evaluated by performance analysis. The most common technique in performance tuning is represented by an even distribution of load over all components. Because the IT environment is changing,
characteristics of load are changing as well. The storage system configuration must be adjusted to reflect these changes. The most common tasks in
performance tuning include selection of the appropriate RAID level, LUN
parameters, front end and back end port configuration, and configuration
of individual data paths. Advanced performance tuning is connected with
cache reconfiguration and the deployment of services such as automated
tiering or wide striping.
What can be measured and what is the output?

Now its time to get more technical. There are three major areas in
our data infrastructure where we can perform measurements and get
the performance data we need for analysis and evaluation. These three
areas (as depicted in Figure 8.2) are:
Hosts and HBA components (Section A)
SAN infrastructure, especially switches and their ports (Section B)
Storage systems and their components (Section C)
278
Host
Switch
4
ILS
2
Back end
Cache
Front end
2
1
Switch
Figure 8.2 This picture depicts all the locations where we can measure performance. Continuous monitoring of these locations provides us with data necessary for performance analysis.
In hosts (servers) we measure read/write response time of mapped

storage volumes. The values we receive are in milliseconds. The response time measured in the hosts provides us with a value that includes both network time (time to transfer the I/O request over the
network) and I/O transaction processing time in storage systems. In
hosts, we also measure the throughput in HBA adaptors (MB/sec), the
IOPS, and the disk queue and queue length. The throughput on HBA
adaptors shows us the amount of data transferred from one particular
host to the SAN. The IOPS shows us how many I/O operations the network and connected storage system can handle per second. The disk
queue and queue length values show you how many pending requests
are waiting for access to the storage system. These values should be as
low as possible. If the disk queue length values are consistently high,
it means the storage system is not able to process all the I/O requests
without delay. The storage system resources must be reconfigured so
that the disk queue length values drop to a low level.
279
St
Gr ora
ou ge
ps Sy
st
e
St
Ba ora
ck ge
En Sy
d ste
Sy
st
e
St
Po ora
rt ge
Co
re
D
h
eS
wi
tc
Ed
g
Ad
a
Ho
st
Bu
s
Ap
pl
ica
tio
n
pt
er
ire
c
to
r
Cache
2
Ex
Sy ter
st na
em l S
to
ra
g
I/O journey starts

and ends at the
application
Figure 8.3 The I/O journey starts and ends at the application. In the host we can measure
the response time, including the sum of time required to process an I/O transaction in each
device located in the data path. The response time measured in the host includes network time
and storage system I/O transaction response time. This measurement can show you there is a
problem with response time, but it does not tell you whether the problem is in the network or
the storage system.
In a SAN we focus mainly on Fibre Channel switches. We can measure throughput on individual ports (MB/sec), buffer credits, cyclic redundancy check (CRC) errors and the operations of interswitch links.
The throughput measured on a particular Fibre Channel switch port
provides us with the amount of data transferred from all hosts connected to this port. Remember that we connect all our hosts to Fibre
Channel switches. These switches are also connected to our storage
systems. Depending on your zoning and infrastructure architecture, the
port linked to a storage system has to service several ports connected
to the host. One Fibre Channel switch port can therefore aggregate
traffic coming from several hosts.
As an example, imagine a situation where three hosts are linked
to one front end port on a storage system. The throughput measured
on the switch should correspond with the throughput measured in
280
the HBA and the storage systems front end port. The Fibre Channel
switches need to be configured so that they are able to handle the
throughput. There need to be reserves in the available bandwidth to
accommodate a peak load without delays or disruptions.
Buffer credit flow control is Fibre Channel switch functionality. Buffer credits represent the number of data frames a port can store. This
is particularly useful for long distance Fibre Channel connections (for
example, a Fibre Channel offsite connection for synchronous remote
replication), where the response time is longer because there are limitations determined by the laws of physics (for example, the response
time cannot be faster than the speed of light). If the network is under a
high load and its bandwidth becomes insufficient, ports can store several data frames that then wait for processing. If a port cannot store
more data frames, the host stops sending I/O requests until space is
freed for more data frames. Buffer credit values can show us whether
our network has any bottlenecks.
CRC and other error reporting mechanisms operate on the switch
level. Their function is to ensure consistency of transmitted data. The
switch communicates with the HBA and front end ports in a storage
system and detects any possible loss of data frames. Data frame loss
can occur mainly if there is a problem with cabling and connectors.
If a data frame is lost during the transport, the host must repeat the
I/O request, which causes delays. This type of error points to problems
with the physical networking infrastructure. All cables and connectors
should be checked.
Note that CRC errors are well known in CD and DVD media, as well as
hard drives. In CD and DVD media, it usually means the surface of the disk
is scratched and the data cannot be reconstructed, which results in missing blocks of data and inconsistency. In a hard drive, the CRC error occurs
when the data is stored on a bad sector. This bad sector is then marked as
damaged and no longer used for data storage.
281
An interswitch link is the connection between individual Fibre Channel switches. The status report shows us whether there are any problems with the connection between individual Fibre Channel switches.
In storage systems we have a number of possibilities regarding what
and where to measure:
Cache Write Pending Rate
- This feature shows us cache is not able to handle all I/O
requests efficiently.
- You must reconfigure cache or upgrade cache capacity.
- Problems are likely to occur when we have insufficient cache
capacity for installed or virtualized capacity (external storage).
Response time on front end ports
Throughput on front end ports
Usage of parity groups
Back end FC-AL or SAS operations
CPU load and load of SAS controller counting parity data
All these values will be described later in this chapter together, with
examples of best practices. In Figure 8.4 you can see an overview of
performance data we can collect. We have mentioned the most important ones, but there are many more. This should give you an idea how
complex performance tuning can be.
282
Storage Systems:
Performance IOPS (read, write, total), MB/sec,

history, forecast and real time monitor
By all storage systems
By port
By processor
By processor
By LDEV
Cache utilization
By disk parity group
By database instance, tablespace, index and more
SAN Switches:
Bytes transmitted/received, frames transmitted/received by SAN, by switch, and by port

CRC errors, link errors, buffer full/input buffer full and more
Servers:
Server capacity/utilization/performance

I/O performance total MB/sec, queue lengths, read/write
IOPS, I/O wait time

File system Space allocated, used, available

Reads/writes, queue lengths
Device file performance and capacity
Applications:
Oracle table space performance and capacity buffer

pools, cache, data blocks read/write, tablespaces used,
free and logs
Microsoft SQL Server cache usage: current cache hit
%/trends, page writes/sec, lazy writes per second, redo
log I/O/second, network: packets sent/received
CPU busy/wait, paging/swapping, process metrics, IPC

Shared memory, semaphores, locks, threads and more
NFS client detail and NFS Server detail
HBA bytes transmitted/received
DB2 table space performance and capacity buffer pools,

cache, data blocks read/write, tablespaces used and free,
and logs
Exchange database, shared memory queue, information
store, mailbox store, public store and Exchange Server
processes
Figure 8.4 An overview of performance data we can collect. Remember that, for performance
analysis and evaluation, you need to collect data over time. Current values are of little use. The
measurements should be performed regularly and data for performance analysis should include
weeks or months of measurement in order to a get precise and helpful image of the utilization
of individual resources.
When you collect all this data, you need to perform analysis. The
analysis helps you investigate the tuning possibilities and suggest configuration changes. It should also help you formulate your long term
performance strategy. Using reporting tools that operate on all the
levels of your infrastructure including hosts, Fibre Channel switches
and storage systems provides you with output data. To be able to
interpret this data meaningfully, you must interpret the meaning of a
value from one source in relation to values from the other sources.
If you have a response I/O time value from a host, you need to interpret it in relation to the I/O response time measured in Fibre Channel
switches and storage systems. To do this you need to analyze the mutual relationship of the values and their meaning, either manually or
using specialized software. See Figure 8.5 for reference.
283
Server
Application
SAN
Storage
Server tool
Switch tool
Storage tool
Server
report
Switch
report
Storage
report
Spreadsheet
Server
Application
SAN
Storage
External
storage
Tuning Manager
Server
Oracle, SQL Server, DB
instances, tablespaces, folie
systems, CPU utilization,
memory, paging, swapping,
file system performance,
capacity, and utilization,
VMguests correlation
Switch
Whole fabric, each switch,
each port, MB/s, frames/sec,
and buffer credits
Storage
Ports, LDEVs, parity group,
cache utilization, performance
IOPS, MB/sec and utilization
Figure 8.5 You can analyze collected values and their relationships manually or by using a
specialized software tool.
Performance tuning in midrange storage systems

In terms of traditional block level storage systems, we distinguish
between midrange and enterprise class monolithic systems. Even
though midrange storage systems boast enterprise level functionalities, their performance and configuration possibilities are still limited
284
in comparison to enterprise class storage systems. Before we discuss

the best practices for midrange tuning management, remember what
a midrange storage system looks like: we have two controllers for redundancy, a limited number of front end Fibre Channel ports, limited
number of hard drives that can be installed and relatively small cache.
Detailed information on midrange storage system architecture can be
found in Chapter 2.
The following examples of best practices in performance tuning are
valid for the Hitachi Adaptable Modular Storage 2000 family. This midrange storage system will serve as an example. Note that performance
tuning best practices for other midrange storage system from various
vendors can differ slightly. In midrange storage systems, we have the
following tuning possibilities:
Controller CPU CPU utilization should be 75% on average. If utilization is higher than 75%, you need to consider whether it is possible
to spread some load to the other controller or other storage system.
Implementation of load balancing functionality can also help to solve
this situation.
Cache Write Pending (CWP) This value can show us whether our
cache is working properly. The cache serves as a buffer that compensates for insufficient performance of physical hard drives. When the
cache gets full, the data must be destaged from cache to individual
disks. This happens asynchronously. When a host sends a write request,
the storage system stores the data to be written on hard drives in cache
and sends acknowledgment to the host, confirming successful write operations. This block is then destaged from cache to the hard drive. The
cache does not send each write transaction immediately to the physical disks. Instead it waits until there are several write requests for the
same destination on the physical disks and then sends all these write
requests to the physical disks together. If the cache gets full and the
physical hard drives have too many pending requests waiting to be destaged to disk to store all the data in cache, we have a few options for
resolutions. We can upgrade the cache if possible, or we can employ
285
cache partitioning, which assigns a portion of cache to the application

causing the high load. These resolutions allow uninterrupted operations of other applications.
When a front end Fibre Channel port is overutilized, we can try to
employ load balancing, or we can manually change paths to other LUNs
so that all front end ports are utilized evenly.
When we take a look at a storage systems back end, the major focus
that affects the performance of the whole system is on individual disks
and parity groups. The utilization of a hard disk unit (HDU) should be
less than 75%. If this limit is exceeded, there is more than a 75% chance
of waiting on a physical I/O request to the back end array group. When
we monitor HDU activity and discover a performance imbalance, we
can try to:
Spread the load to another RAID group
Change the RAID level to one more appropriate for the workload
Use faster disks
Add more disks to the RAID group
Move some of the load to the other storage system
The LUN response time is also monitored. In a Microsoft SQL Server
environment, the LUN read response time for the database disk should
be below 20ms on average, and spikes should not be higher than 50ms.
The response time of a particular LUN can be evaluated as follows:
Less than 10ms very good
Between 10ms and 20ms okay
Between 20ms and 50ms slow, needs attention
Greater than 50ms serious I/O bottleneck
286
Enterprise storage performance basics

Compared to midrange storage systems, enterprise level systems
are much more robust and powerful. They offer a significantly higher number of front end ports, extremely large cache and their back
end architecture allows connection of many expansion units with hard
drives. This robust architecture provides many more possibilities for
performance tuning. Performance tuning in enterprise storage systems
can be very complicated because of configuration complexity. Lets take
a closer look at individual areas of an enterprise level storage system
that are particularly interesting from the perspective of performance
tuning:
Response time factors
Port utilization and throughput
Cache utilization
Response time is affected by utilization of the particular front end
Fibre Channel port, cache and individual parity groups. These are the
hardware limitations. However, response time is also affected by load
characteristics. It really depends on whether we are using applications
that store data sequentially or applications that request random I/O.
As we discussed previously, traditional enterprise storage systems work
on the block level (as opposed to the file or object level). The size of
blocks is also important. Lets look at the operations associated with
enterprise level front end Fibre Channel ports in more detail.
Port load balancing is important and should be employed in an enterprise level storage system. A single physical front end Fibre Channel
port is able to process thousands of I/O requests. It is also possible to
create virtual front end ports to prioritize the traffic to certain hosts.
Front end ports need time to process protocol instructions and virtual port operations; this process is known as performance overhead.
In contemporary storage systems, the overhead associated with protocol time on a physical port is very low. It is currently calculated in
287
microseconds, where it previously was calculated in milliseconds. The

performance overhead still needs to be considered, especially when
implementing a high number of virtual ports per physical port. Block
size is critical for planning and analysis. A single Fibre Channel port
can process thousands of I/O requests before reaching the saturation
point. This point is reached faster in cases where the average block size
is large (64K and higher). Performance problems would be seen when
load balancing is not performed properly during configuration and/
or capacity planning. Performance problems could occur if too many
virtual ports are created on a single physical port or if there are too
many active LUNs on the same physical ports. Front end port virtualization and port priority assignment are configured on the storage system
level.
With port utilization and throughput optimization, you may still face
a few problems. If you have a high number of IOPS with small blocks,
the port capacity is saturated later than if you have a smaller number of IOPS with large blocks. However, a high number of IOPS with
small blocks stresses the CPU and creates a performance bottleneck.
You need to carefully plan how many hosts you will connect through a
single Fibre Channel port, how many virtual ports you will use and what
type of load the port will deal with. It is important to find a balance to
avoid the problems connected with lack of performance.
Cache utilization is a key element for great I/O performance. Remember that you want to avoid CPU and hard drive utilization higher
than 75%. In cache, you want to do the exact opposite. The more cache
you utilize, the better. The cache utilization target is 90% or more. As
for the cache sizing, remember that a larger cache is important, especially for random I/O. A large cache permits a frame to remain in cache
longer, which increases the chance of a read hit. If you send the frequently used data into a storage system with a large cache, it can happen that this data remains in the cache until the server sends a request
to read this particular data. In other words, if you have a large cache,
frequently used data can stay there for repeated reading of data. A high
read hit rate is desirable because the storage system can handle the
288
read request much faster when the desired data is stored in cache than
if this data has already been destaged to hard drives. You can imagine
this as the highest tier when frequently used data stays in cache, it
can be accessed much faster than if it was stored on a hard disk or even
a solid state disk. Note that by frequently accessed data we mean data
that is accessed several times a minute. It is therefore a very small portion of data compared to the total capacity of a storage system.
When it comes to cache configuration, another thing to remember
is that sequential workloads use smaller amounts of cache memory as
compared to random workloads. The key to success, therefore, is in
understanding how an application utilizes cache. Applications have different cache utilization characteristics.
Impact of cache size on performance
A positive impact will be seen using larger cache, especially on workloads that are predominantly random read/write activity. The larger the
cache is, the longer a frame will stay in cache, thus increasing the chance
of a read hit on the same frame. Sequential workloads use small amounts
of cache memory and free cache segments as soon as the job is completed. As a result, cache sizing generally has little to no impact on sequential
workloads.
The size of cache you should use, therefore, largely depends on how
your application works. However, there are some basic rules for cache
sizing. When you upgrade your infrastructure with a new storage system, you should use at least as much cache as the storage system being replaced. Increase cache proportionally to any increase in storage
capacity. If you do not have any current performance data to start with,
configure at least 1.5GB to 2GB of cache per TB of installed capacity.
Start with the current hit rate if you know it. Remember the guideline
that doubling cache size reduces the miss rate by half. For example, if
your read hit rate is 80% with 16GB of cache, then doubling cache size
to 32GB should improve hit rate to 90%.
289
The last thing you need to consider when configuring cache is what
portion of cache will be used either for read or read/write operations.
The correct settings of cache proportions help to achieve higher performance. If these settings are not correct, the write and read operations
may perform more slowly than desired.
In addition to ports and cache, we can optimize LUN response time
and parity group performance. The LUN response time characteristics
and optimization are practically the same as in midrange storage systems and have been discussed earlier in this chapter. Parity group performance optimization is, again, similar to the midrange storage systems, so lets just look at a few of the most important tips:
Treat every sequential I/O request as four random I/O requests
for sizing purposes.
Do not allocate more than four hot LUNs per parity group. The
LUN is characterized as hot when its utilization reaches 70% or
more.
Do not allocate more than eight medium hot LUNs per parity
group. The LUN is characterized as medium hot when its utilization ranges between 50% and 70%.
Remember that to tune your storage system to the highest possible

performance, you need to consider not only the hardware capabilities and
the evenness of the workload distribution, but also applications that access the storage system. You should know whether your applications access data sequentially or randomly. The data block size is also important.
Please refer to Figure 8.6 for a clear and straightforward enumeration
of the most important performance tuning metrics.
290
Metric Name
I/O rate
Read rate
Write rate
Read block size
Write block
size
Read response
time
Write response
time
Avarage queue
length
Read hit ratio
Write hit ratio
Average write
pending
Value Description
I/Os per second
I/Os per second
I/Os per second
Bytes transferred per
I/O operation
Bytes transferred per
I/O operation
Time required to
complete a read I/O
(millisecond)
Time required to
complete a write I/O
(millisecond)
Average number of
disk requests queued
for execution on one
specific LUN
% of read I/Os satisfied
from cache
% of write I/Os satisfied
from cache
% of the cache used for
write pending
Normal Value
N/A
N/A
N/A
Bad Value
N/A
N/A
N/A
4096
N/A
4096
N/A
10 to 20 (DB)
2 to 5 (log)
>20
>5
10 to 20 (DB)
5 to 10 (log)
>20
>10
1 to 8
8 is the maximum queue
depth value
>8
25% to 50%
<25%
100% (0% on Hitachi

Virtual Storage Platform)
<100%
1% to 35%
>35%
Figure 8.6 Overview of metrics relevant to performance tuning and recommended normal
and bad values.
Software for performance tuning and automated

optimization
Tuning management software is a storage performance management application that maps, monitors and analyzes storage network
resources from the application to the storage device. It provides endto-end visibility you need to identify, isolate and diagnose performance
bottlenecks with a focus on business applications, such as:
Oracle
Microsoft SQL Server
291
Microsoft Exchange
IBM DB2
This software correlates and analyzes storage resources with servers
and applications to improve overall system performance. It continuously monitors comprehensive storage performance metrics to reduce delay or downtime caused by performance issues. It facilitates root cause
analysis to enable administrators to efficiently identify and isolate performance bottlenecks. It allows users to configure alerts for early notification when performance or capacity thresholds have been exceeded.
In addition, it forecasts future storage capacity and performance requirements to minimize unnecessary infrastructure purchases.
Among other features of advanced tuning management software,
we can find tools that provide in-depth performance statistics of storage systems and all network resources on the applications data path. It
works together with storage functionalities such as thin provisioning or
automated tiering pools for usage analysis and optimization.
This software provides customizable storage performance reports
and alerts for different audiences and reporting needs.
When we implement such software, we can benefit from:
Simplification of performance reporting and management of the
storage environment
Improvement of service quality through accurate performance
reporting
Increased application availability through rapid problem identification and isolation
Reduction of storage costs through proper forecasting and planning of required storage resources
292
Automation of performance tuning is usually available only on enterprise level storage systems. Prerequisites for such automation are
implementation of thin provisioning and automated tiering. Some storage systems have utilities for performance monitoring and tuning inbuilt in the microcode.
Good tuning management software focuses not only on storage
systems but also on the SAN and hosts. It can happen that we look
for problems in the storage systems when the problems are actually
caused by incorrect application settings.
Consolidation when and why?

The issues companies face nowadays are mostly connected with
a heterogeneous IT infrastructure and the complexity of its management. The consolidation should address these issues and bring us to a
lower TCO. The consolidation provides a long term effect; while we do
not save money immediately after the consolidation of an IT infrastructure, we will see savings in the following years. The target of consolidation is to:
Centralize and simplify management
Save electricity and space
Reduce complexity and heterogeneity of IT environment
Reduce number of personnel that perform administration and
main-tenance
Save money on licensing, as well as personnel training and certification
Unify and standardize hardware platforms and hardware operations
293
The means that allow consolidation include blade servers, server

virtualization, modern storage systems with lower electricity consumption, storage system virtualization, outsourcing to large data centers
and cloud computing. Effective equipment utilization, which is connected with performance tuning, also helps to lower the costs.
We should consolidate our IT infrastructure when we detect increasing costs related to storage system operations and maintenance. We
consolidate servers, network devices, storage systems and applications.
Consolidation project
When we realize that our company is spending too much money on
IT infrastructure operations, we should take action. The consolidation
starts with a consolidation project, which can be divided into four stages. To illustrate how a consolidation project looks and what it should
include, we will use consolidation of servers and storage systems as an
example situation. The stages of a consolidation project are:
Analysis of the current IT infrastructure state
Development of consolidation plan
Migration to new consolidated platform
Optimization of the new solution
The first step is to analyze the current IT infrastructure state. To do
so, you need to write down all your hardware and software resources.
Among hardware resources there are servers, networking components,
storage systems and workstations. Software resources include operating systems and applications. When you complete this list, you need to
focus on how all these resources work together and what performance
they deliver. You need extensive, detailed information on how your
hardware resources are implemented, interconnected and secured. As
294
for the software resources, you need to know the versions of the operating systems and applications are being used, a history of updates
and patches, and a detailed licensing analysis. You could use performance reports together with the performance monitoring history. Bills
for electricity and personnel training may come in handy as well.
Essentially, we can say that the first stage of consolidation, the analysis stage, focuses on three levels: technical, financial and performance
statistics. You can see that there are plenty of factors that need to be
considered. Because of the consolidation projects complexity, it is not
something that can be completed overnight. It can take up to several
months and the efforts of both IT management and IT administrators.
In the second stage of IT infrastructure consolidation, we take the
output of the analysis we made in the first stage and suggest a solution. There are many strategies on how to do this, but the key is to have
comprehensive knowledge of the products available on the market
that could help in consolidation. We also need to calculate operating
expense savings to prove the profitability of the new solution. Nowadays, most consolidation projects include migration to blade servers
and virtualization. Virtualization in servers is usually much simpler and
cheaper to implement than virtualization of storage systems.
The third stage is the migration to a new consolidated platform. To
perform a seamless migration, we should stick to the well known Information Technology Infrastructure Library (ITIL) methodology. Migration can be physical to virtual (P2V). Virtual to virtual (V2V) migration is
an option when we are talking about consolidation in an environment
that has already been virtualized.
Once we have migrated to the new solution, the last stage takes
place. We perform regular monitoring to determine whether the expected results of consolidation have been reached.
In this chapter, you learned performance tuning basics. You are now
familiar with basic performance metrics. You know where in your IT equipment you can make performance measurements, and you are able to interpret the output of these measurements. We have also discussed what
software tools you can use for performance tuning and what these tools
usually offer. Remember that the values presented in this chapter represent general information. Recommended performance values for your
storage system can differ from what is mentioned here. Please refer to
the documentation that was supplied with your storage system for more
detail.
295
296
Business challenges
How to see a storage systems infrastructure from the
perspective of IT managers
What business challenges companies face
How to accommodate accelerating data growth
How to describe advanced data classification and tiered
storage
Environmental concerns connected with data center
operations
The importance of TCO, CAPEX and OPEX calculations
298
9 Business challenges
Business challenges
Previous chapters of this book dealt mostly with the technical aspects of storage systems and SANs. You also gained insight into storage
system administration and maintenance. In this final chapter, we will
provide you with yet another perspective on storage systems. We will
show you what challenges companies face today and what solutions
help IT managers meet these challenges. This chapter will therefore
be particularly useful to IT managers. The data provided in this chapter
originated from complex research conducted by Hitachi Data Systems
on its customers. You will learn what the most painful issues are and
the options we have to overcome them. You will also be introduced to
some functionalities of storage systems that are designed to be environmentally friendly.
At the beginning, its necessary to realize that different companies
have different needs and they encounter different challenges. Your organization may face extremely fast data growth and need more and
more capacity to accommodate such growth. Another possibility is that
your organization does not suffer from a lack of storage capacity, but instead has serious problems with high availability this happens when
your data becomes so important that you cannot afford any disruptions. You may also have problems with a lack of physical space to place
new racks, insufficient cooling systems unable to service newly integrated devices and other issues. As a result, we will need some clear
classification of business challenges to avoid getting lost in the myriad
of possibilities. We can classify the challenges as follows:
Accelerating storage growth need more and more capacity
Increasing requirements on high availability cannot afford
any dis-ruptions because our data has become too important for
our business
Fast and effective response to business growth need to be
able react quickly to new conditions
299
Heterogeneous infrastructure result of fast infrastructure

growth, which was not properly planned, causing TCO to increase
rapidly
Compliance and security challenges need to process, protect
and retain data according to legal requirements and regulations
Power and cooling challenges pay too much for electricity and
cooling and may be running out of space in the server room
Data center challenges specific needs for those whose business is focused primarily on cloud type provision of storage and
computing capacity
In the following subsections of this book, we will address each issue.
Data growth forecast

To be able to keep up with the storage capacity requirements, you
need to perform data growth analysis and implement capacity management. You need to gather information about capacity requirements
of all the applications and users, and forecast how these requirements
will grow in the horizon of the next months and years. To be able to
forecast storage capacity for the future, you need to analyze and evaluate data from the past. To make a prediction, you need to know whether the data growth has been linear or geometrical. You need to have
clear statistics on capacity requirements for storage tiers as well. The
demand for high performance data storage can grow more quickly than
the demand for low performance storage, or vice versa. You also need
to consider requirements on RAID protection and data replication because data must be stored in several copies for redundancy and other
purposes.
300
300
Capacity (GB)
250
200
150
100
50
0
2008
2009
2010
2011
2012
2013
Figure 9.1 An overview of the storage requirements in the past years provide you with the
necessary information to forecast data growth.
The demand for storage capacity can be linear. If your storage infrastructure has not been consolidated yet, it can happen that the amount
of stored data drops once you centralize and consolidate your storage.
During consolidation, you also classify data and implement tiering. A
portion of your data can be archived to tapes, freeing space for live
data. Linear data growth is represented by a straight line in a graph. If
the growth is linear, you can forecast growth easily.
However, more often, the demand for storage capacity is geometrical, meaning the amount of data coming into existence is growing exponentially. Exponential growth is often caused by an increase of unstructured data, large quantities of metadata or an increased demand
for in-system replication, where S-VOLs are used by data mining or
warehousing applications.
301
Lets take a look at some example data growth forecasts:

Linear growth our analysis showed that each year the amount of
stored data increased by 30TB. In the horizon of five years, we will need
150TB of capacity.
Geometrical (exponential) growth our analysis showed that each
year the amount of stored data increased by 120% compared to the previous year. The growth is depicted as a curve on the graph.
To achieve economical utilization of all resources, we also need to

anticipate what kind of data will grow in what amount. Structured data
and unstructured data have different hardware requirements. Structured data, represented mainly by databases, should be stored on a
suitable storage system. Traditional block level storage systems offer
high performance in processing transaction based data. Conversely, unstructured data (such as pictures and PDF files), are best processed and
organized by NAS devices and digital archives. We need to anticipate
how we will grow both structured and unstructured data to plan development of our storage systems accordingly. For example, if we know
the amount of unstructured data will grow much more quickly than the
amount of structured data, we can plan a capacity and performance
upgrade in the NAS device or digital archive.
We also need to perform an analysis of tier utilization to learn what
data requires what performance. Depending on our analysis, we may
learn that data saved on high performance tiers will grow much more
quickly than data stored on low performance tiers, or vice versa. See
Figure 9.2 and Figure 9.3 for reference.
302
Data growth forecast - tier view

350 000
300 000
Capacity (GB)
250 000
200 000
Tier 1
150 000
Tier 2
100 000
Tier 3
50 000
0
2011
2012
2013
Years
Figure 9.2 Data growth forecast in relation to performance tiers.
250 000
Capacity (GB)
200 000
150 000
Replicated data
100 000
Unstructured data
Traditional
Structured data
50 000
0
2005
2006
2007
2008
2009
2010
2011
Years
Figure 9.3 Structured, unstructured and replicated data growth.
303
Advanced data classification and tiering

In Chapter 5, we discussed the automated tiering functionality,
which is offered in modern enterprise level storage systems. Automated tiering uses a virtualized storage pool, which is spread across several
tiers. One example would be parity groups using hard drives with different performance and price characteristics. The storage system then
stores data automatically on individual tiers according to the frequency
of access. This functionality offers great performance and allows you to
achieve high utilization of expensive resources. However, to economically use all your storage resources, you need to design and implement
tiered storage systems that distribute data to individual tiers according to many criteria not only frequency of access. Tiered storage
(not to be confused with automated tiering functionality) represents
the whole infrastructure of storage systems, where all resources are
effectively utilized and where data is stored on hard drives according
to its requirement on availability, security and performance. See Figure
9.4 for reference.
Storage Service Level
Expensive
Inexpensive
Performance
Tier 1
Tier 1
Tier 2
Tier 1
Tier 1
Tier 2
Sec
uri
ty
ai
Av
lit
i
ab
Tier 2
Tier 2
Tier 3
Tier 3
Figure 9.4 Tiered storage infrastructure.
304
Implementation of a tiered storage infrastructure is a very powerful solution, which can offer many benefits, especially in terms of TCO.
Implementation of a tiered storage infrastructure would, however, be
useless without careful classification of data. This is the task of IT managers who are responsible for IT infrastructure design, development
and maintenance, including a SAN. Data classification is primarily a
manual process that cannot be fully automated. It is the IT manager
who has to discover and evaluate information assets and determine
whether or not this information has worth for your business. Data classification is, therefore, not only based on frequency of access, but also
on data value. For reference, visit the Storage Networking Industry Association (SNIA) Data Management Forum at www.snia.org. SNIA defines information classes in the following categories:
Mission-critical
Business vital
Business important
Important productive
Not important
Discard
At this point, its good to realize that we also need to determine
what kind of data can be deleted. Companies and their employees are
reluctant to erase data, but there should be certain policies governing
data shredding as well. These should be included in the information
lifecycle planning. To assign data to one of the above mentioned categories, we can use various metrics:
Usage pattern of the information frequent, regular, periodic,
occasional, rare, on demand or request, or never
Information availability requirement immediate, reasonable,
defined time frame, extended time frames, limited, not defined
or unnecessary
305
Financial impact of information unavailability significant longterm and/or short-term, potential long-term, possible, unlikely or
none
Operational impact of information unavailability significant
and immediate, significant over time, probable over time, possible over time, doubtful or none
Compliance impact of information unavailability definite and
significant, eventual, probable, potential or possible, or none expected
Tier 1 High Performance
Mission Critical Applications
SCM
ERP
Tier 1 Good
Performance
Business
Critical
Applications
Data
Mining
E-mail
Hitachi
Universal
Storage Platform
CAD
IBM Enterprise
Storage Server
Data
Warehouse
EMC
DMX2000
Hitachi Virtual
Storage Platform
CRM
NAS
Video
Tier 4 Low Priority

Non Critical
Print
Nearline or Offline
Storage
CLARION
Online
Tape
Hitachi
Adaptable
Modular
Storage 2300
Tier 2 Medium Performance

Non Critical Applications
Hitachi
Adaptable
Modular
Storage 2100
Tier 3 Low Priority Non

Critical Applications
Figure 9.5 Overview of virtualized multitiered storage infrastructure with implemented data
classification.
As you can see, plenty of criteria should be considered. Assessing

data importance and classification, together with maintaining policies in this respect, is time consuming and resource intensive. For this
reason, IT managers often neglect data classification and avoid implementing a multitiered storage approach, but this decision results in unnecessarily high TCO.
306
Data classification is the responsibility of IT managers, and it results in

the division of data among data groups that require specific handling. Each
company can use different data classification procedures. The goal is to
assess the value of information for a particular business. According to the
data groups, we define Storage Service Levels (SSL). SSLs contain detailed
information about hardware resources. New equipment is purchased in
accordance with SSLs.
Data classification is a prerequisite to implementation of multitiered
storage infrastructure. As you can see in Figure 9.4 and Figure 9.5, a multitiered storage infrastructure represents a whole solution that includes
implementation of different kinds of storage systems. In a multitiered
storage infrastructure, we can have enterprise level storage systems, midrange storage systems, NAS devices, digital archives and tape libraries.
A multitiered storage infrastructure is a powerful and complex solution. Implementation of this solution can be the result of storage infrastructure consolidation.
A multitiered storage infrastructure is different from automated storage tiering because it considers various criteria, not only frequency of access. Virtualization and thin provisioning are often part of capacity and tier
management.
307
Throughput
Enterprises design
Host multipathing
Hardware load balancing
Dynamic Optimized Performance
Dense storage, SAS, and SATA
99.999+% reliability
e
lac
n-p e
i
ta
ad
Da pgr
U
Adaptable Modular
Storage 2100
4-8GB Cache
4FC, or 8 FC, or 4 FC and
4iSCSI ports
Mix Up to 159 SSD, SAS
and SATA-II Disks
Up to 313TBs
Up to 2048 LUNs
Up to 1024 Hosts
e
lac
n-p e
i
ta
ad
Da pgr
U
Adaptable Modular
Storage 2300
8-16GB Cache
8FC, or 16 FC, or 8 FC and
4iSCSI ports
and SATA-II Disks
Up to 472TBs
Up to 4096 LUNs
Up to 2048 Hosts
Adaptable Modular
Storage 2500
16-32GB Cache
16FC, or 8 iSCSI, or 8 FC and
4iSCSI ports
and SATA-II Disks
Up to 492TBs
Up to 4096 LUNs
Up to 2048 Hosts
Scalability
Figure 9.6 Not all organization are big enough to implement a multitiered storage infrastructure. For small and midsized organizations and companies, there are modular storage systems designed to meet their needs. In this picture, you can see the Hitachi Adaptable Modular
Storage 2000 family and its scalability. This scalability and the possibility of easy upgrades are
important features that should be considered when designing a storage system infrastructure.
High availability challenges

When your data becomes so important that your business cannot
have any service disruption, your major concerns are protection against
data loss and business operation continuity.
The solution can be found in the advanced replication and backup
functionalities offered in enterprise level storage systems. Modern enterprise level storage systems offer unified out-of-region data protection that ensures the availability of up-to-date data replicas. To deliver
308
the best protection available, you can implement three data center
multitarget replication. Please refer to Chapter 4 for more details. The
benefits of such a solution lie in fast data synchronization with long
distance recoverability and in reduced storage cost by dynamically provisioning pools at backup sites.
The universal replication feature, which is available in modern enterprise storage systems, supports synchronous and asynchronous replication for both internal and external virtualized storage. A single replication framework for all externally attached arrays by different vendors
represents a powerful solution that can effectively address all issues
connected with high availability clustering. There is no requirement
to implement multiple, disparate business continuity solutions across
each array. This significantly minimizes costs, reduces infrastructure
and simplifies disaster recovery management and implementation.
This single framework approach to managing business continuity also
eases IT manager concerns related to having a consistent, repeatable,
enterprise wide disaster recovery process.
Virtualization of external storage also simplifies backup solutions
since the backup infrastructure only needs to be zoned to one storage
system, no matter how many different physical storage systems are virtualized behind it. This simplifies the overall management of the entire
centralized backup infrastructure.
Fast and cost-effective response to business growth

The response time to meet business needs for additional storage
and the cost of provisioning for unplanned data storage are business
challenges as well. When our capacity planning fails and we unexpectedly need to purchase more capacity, we realize this problem can be
very painful.
309
We can solve this issue by implementing thin provisioning, which

speeds the availability of additional storage and reduces the risk of having insufficient storage. Benefits of thin provisioning include improvement of access to information, increased productivity and risk mitigation and elimination of costs associated with not meeting fast changing
business needs.
Thin provisioning functionality is available both on internal and external virtualized storage. It uses storage pools and virtual volumes.
Through thin provisioning, the storage system administrator can assign
virtual capacity to hosts even though the physical capacity is not yet installed. Virtual volumes are mapped to servers and capacity is provided
by storage pools. The storage system monitors storage pool usage continuously and sends alerts to the administrator when it is necessary to
install more physical capacity. The important feature is that capacity
expansion can be done nondisruptively.
Load balancing and active-active architecture maximize performance without allowing the actions of one set of users to affect the
quality of service of others, thus, readily accommodating new classes
of users and applications.
Nondisruptive data migration capabilities are also critical. In-system
replication able to create volume clones across externally connected
storage systems ensures data migration is as simple as it can be. When
you create a volume clone, you do not encounter decreased performance or service disruption.
Compliance and security challenges

With new regulations and legal requirements on data processing,
protection and retention, companies are struggling to comply and cut
risks. A solution that addresses these challenges should be capable of
310
blocking unauthorized access and protecting data with no performance

impact.
As high profile data indiscretions continue, the response is tighter
data security regulations to control sensitive information. Increasingly,
these regulations are being enforced to protect sensitive data, and the
risks associated with noncompliance to these tough regulations are significant. Organizations and companies are employing all prudent measures at their disposal. The storage level data encryption feature is one
such measure.
Modern high-end storage systems are available with a controller
based data-at-rest encryption option compatible with both open and
mainframe systems. This encryption can be used to encrypt the internal drives, using strong encryption (AES-256). In modern powerful storage systems, this encryption introduces absolutely no throughput or
latency impact for data I/O. The feature also includes an integrated key
management functionality, which is both simple and safe.
Beyond the cryptographic characteristics, the data-at-rest encryption feature can be used to introduce encryption within a storage ecosystem with little to no disruption to the existing applications, infrastructure or network.
This feature has the added benefits of being data center friendly
because it uses very little additional power (approximately the equivalent of a 25W lightbulb for 1PB of data), produces negligible amounts
of additional heat and requires no additional rack space. Other benefits
include reduced complexity and decreased encryption cost.
311
Power requirements and cooling

As requirements for more and more storage space rise, IT managers are increasingly worried about the cost of electricity and cooling.
Nowadays, costs related to electricity consumption and cooling requirements are among the top three issues.
What is the greatest facility problem

with your primary data center?
29%
Power and
Spending
(US$B)
29%
New serve
Power and
$80
$70
21%
$60
13%
$50
6%
$40
3%
$30
$20
Excessive
Heat
cility problem
center?
Insufficient
Power
Insufficient None of the Poor

Rased Floor above
Location
$10
$0
$80
$70
$60
(M units)
New server spending

Power and cooling
of the Poor
Location
3%
14
$30
6
4
$10
$0
16
10
$40
$20
Excess
Facility Cost
18
12
$50
6%
1996 1997 1998 1999 2
Power and cooling exceeds server

spending
Spending
Installed base
(US$B)
Excess
Facility Cost
2
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Figure 9.7 Greatest issues organizations and companies face.
312
Electricity and cooling costs are not the only motivation for reducing
power consumption and implementing electricity saving features. The
onset of global warming is expected to increase the demand on existing
cooling systems, leading to costly investments in new plants. More efficient use of an existing storage infrastructure can reduce the demand
for new storage, ease the demand for power and cooling and enable
a business to reduce costs and address its green obligations. Some of
the challenges facing data centers in respect to electricity, cooling and
environmental requirements include:
Running out of power, cooling and space
Growing energy costs
Increasing regulatory compliance issues
Data center expansion without consideration for future power
and cooling requirements
Data storage configured without adequate consideration to heat
distribution (equipment racks should be installed with cold rows
and hot rows this will be discussed later in this chapter)
Difficulty relieving data center hot spots without disrupting applications
The solution capable of addressing these issues should provide features that lead to cost-effective and energy efficient management and
operation of storage systems. It should also provide the ability to extend the life of existing storage assets.
When you design a storage infrastructure, you need to analyze the
power and cooling requirements as well as the requirements on the
space in your server room. In addition to power consumption metrics
in kW, there are other metrics that should be considered:
Total five-year power and space costs
Heat loading (kW/sq ft)
313
Space requirements (sq ft)

Floor loading (lbs/sq ft)
Aside from these metrics, you should look for a vendor that designs,
manufactures and supports environmentally friendly storage infrastructures throughout their entire lifecycle including end-of-life disposal.
You should also keep in mind that innovations such as controller-based
virtualization and thin provisioning yield substantial environmental advantages because they reduce the need for storage capacity.
Virtualization and consolidation of hardware platforms are the key to
reduced power and cooling costs. Without virtualization, data centers may
be stuck with existing physical configurations leading to an increased need
for cooling. Virtualization strategies must embrace as much of the storage
infrastructure as possible to optimize utilization and save on power and
cooling. Virtualization must be easy to manage through one common user
interface, technology and application.
Consolidation is important because disparate storage is inefficient. Assets are typically underutilized, unnecessarily consuming power, cooling
and space resources. Planning future storage capacity and performance
avoids hitting any limitations that would impact the environmental savings
model.
HDD and fan power savings

Thin provisioning and virtualization are advanced features that affect the management and operations of a whole storage infrastructure.
Their implementation is reflected in lower TCO. However, there are features that work on the HDD and fan level and lead to significant power
savings.
Modern storage systems, both modular and enterprise are in most cases able to spin down drives in selected RAID groups. When properly con-
314
figured, SATA drives will park heads when idle for more than 2hrs. Modern
storage systems must also constantly assess cooling requirements and dynamically adjust fan speeds to maintain correct temperatures.
It would be useless to implement a spin down feature on RAID
groups that are highly utilized (production volumes used by transaction
based applications). However, there are disk functions eligible for spin
down solutions. These include:
Mirrored drive groups involved in backup to tape
Virtual tape library drive groups involved in backups
Local (internal) backup copies of data
Drive groups within archive storage
Unused drive groups
The capability to turn off hard drives after they are not accessed for
2hrs can increase power savings by 23%. This functionality is often implemented together with smart algorithms that minimize HDD usage.
These algorithms ensure that sequential writes and reads are done in
massive batches. Data to be written to a particular hard drive is kept
in cache as long as possible. Imagine that within a few hours there are
only four write requests for that particular hard drive. The algorithm
ensures these four write requests are sent to the hard drive in a batch
instead of accessing the hard drive four times.
The storage system controller also varies the fan speed to the internal temperatures to reduce power requirements for fan operation.
Green data center

An environmentally aware data center that meets requirements on
performance, capacity and high availability, all while remaining eco
315
friendly and power efficient, is now possible. In 2009, this ideal data
center was built and began its operations. See Figure 9.8 for reference.
Applies principles from Eco friendy Data Center Project and IT Power Saving Plan
Location: Yokohama
Operation: 2009
Dimension: 10,000m2
Power Usage
effectiveness rating: 1.6
Rooftop forestation
Air conditioning facility
Electricity generator
High voltage substation
Revolving gate
High pressure disk
Uninterruptible power system
Photovoltaic light
Nitrogen gas extinguishing system
Security camera
Infrared sensor
Integrated surveillance center

Seismic isolator
Figure 9.8 Environmentally friendly and energy efficient data center.
In large data centers, racks with storage systems and servers are
placed in rows. Modern storage systems are designed for such positioning. The goal is clever distribution of heat by managing air flow. As
a result, you get hot rows and cold rows. In simple words, you arrange
racks in alternating rows with cold air intakes facing one way and hot
air exhausts facing the other. Typically, the front side of a storage system is designed to face the cold row, where cold air is distributed from
the air conditioner. The rear side of a storage system is then pouring
heated exhausts into hot rows. Typically, hot rows face air conditioner
return ducts, which take the hot air away.
Figure 9.9 The concept of hot and cold

rows.
316
Total cost of ownership

Throughout the book, we have encountered the term Total Cost of
Ownership (TCO) many times, and we have explained it briefly. You
learned that the initial cost of a new storage system or server is only
the beginning of a long succession of other related costs. TCO is, therefore, a focus for all IT managers and provides the basis for the storage
economics field. Each IT manager needs to carefully calculate TCO for
each device to decide whether it is an economical purchase. Vendors
often use TCO values in their marketing to persuade companies to buy
their products. At this point, it is good to take a closer look at TCO and
list all the items that constitute it. TCO includes all costs related to a
particular device throughout its lifecycle from the purchase to endof-life disposal. TCO is important for budgeting IT resources and for decision making related to old technology replacement.
Floor space
Power and cooling
SLA penalties: backups
Scheduled downtime
Outage risk
Contract labor
Internal labor
Hardware maintenance
Software maintenance
Purchase
Figure 9.10 Items included in baseline TCO.
317
Throughout this chapter, we have discussed many business challenges. Most of the challenges presented are related to TCO, and most
of the solutions presented help to lower TCO.
Common TCO items include hardware acquisition costs, hardware

maintenance, software licensing costs, software maintenance, storage
management labor, backup, migration and remastering, power and cooling costs, floor space, and SAN resources.
In addition to considering TCO, an IT manager must consider distribution of costs over time, such as the costs at purchase, the costs
after one year and the costs after five years. These expenses are calculated through two other values capital expenditure (CAPEX) and
operational expenditure (OPEX). CAPEX represents mostly the initial
purchase and the investments into technology. In other words, CAPEX
is how much you have to spend to get new equipment, licenses and
personnel. OPEX represents the costs for maintenance, electricity and
cooling.
CAPEX can be lowered by implementing functionalities such as thin
provisioning and zero page reclamation. OPEX can be lowered by virtualization and consolidation of infrastructure through higher utilization
of hardware resources.
The motto for lowering TCO, CAPEX and OPEX can be best described
as: Reduce costs by implementing technology! By now you have learned
that virtualizing and consolidating storage infrastructure can result in major cost savings. Implementing data tiers is also very important. All these
factors represent major investments, but when properly calculated, they
will pay you back through long term cost savings. Companies are often reluctant to invest large quantities of money into storage infrastructure, but
the financial figures are clear it is not efficient to maintain old technology that is no longer supported. From a long-term perspective, it is always
best to keep your hardware up to date.
318
320
1.
In the context of storage systems, a volume is a:

a) Logical interface used by an operating system to access data
stored on a particular medium
b) Physical hard drive
c) Record containing information on partitioning of the disk
d) Unique identification number of a partition
2.
In a disk array, data striping is a functionality that is:

a) Controller based and ensures the data is mirrored to two disks
b) Controller based; the controller distributes data over several
physical hard drives to enhance storage system performance
c) Server based; the file system distributes data into stripes according to frequency of access
d) Controller based and ensures the organization of data in the
storage systems cache
3.
The RAID functionality is important because it:

a) Provides economical storage of data on hard drives
b) Allows to create several logical disks over a single physical
hard drive
c) Redundancy feature that prevents data loss in case one physical hard drive fails
d) Access data stored on hard drives through TCP/IP protocol
4.
321
The most important components of a midrange storage system

are:
a) Controller, switch and expansion units with hard drives
b) CPU, hard drives and RAID controller
c) Four controllers for redundancy, cache and SAS back end
d) Front end ports, cache, back end and hard drives
5.
To connect a server to a SAN, we need:

a) HBA installed in the server, Fibre Channel cable and Fibre
Channel switch with available ports
b) Twisted pair cable with RJ45 connector
c) To configure storage system and register server WWN
d) Fibre Channel switch
6.
The implementation of zoning in SAN environment helps to:

a) Ensure QoS for I/O operations
b) Secure SAN network
c) Distribute load evenly over two or more data paths in SAN
d) Interconnect two Fibre Channel switches
322
7.
The main difference between a SAN and NAS is that:

a) NAS works on the block level with a TCP/IP connection, while
a SAN works on the file level with a Fibre Channel connection
b) NAS works on the file level with a TCP/IP connection, while a
SAN works on the block level with a Fibre Channel connection
c) NAS works on the file level with a Fibre Channel connection,
while a SAN works on the block level with a TCP/IP connection
d) NAS works on the block level with a Fibre Channel connection,
while SAN works on the file level with a TCP/IP connection
8.
The purpose of business continuity planning is to:

a) Minimize the negative impact of anticipated threats on a business process
b) Create a disaster recovery plan with step by step instructions
for recovery
c) Optimize the total cost of ownership
d) Describe remote replication requirements
9.
A backup strategy to support 99.999% availability includes implementation of:

a) Advanced backup on tape libraries
b) Snapshots, in-system replication and offsite tapes
c) Remote replication and geoclusters
d) RAID 6, which prevents data loss in the event of failure of two
physical hard drives
323
10. Implementation of thin provisioning allows us to:

a) Share LUNs among several servers
b) Access the hard drive over iSCSI
c) Access data stored on S-VOLs (secondary volumes of a replication pair)
d) Allocate more disk space than is actually installed in a storage
system
11. Storage virtualization:

a) Enables you to view an entire SAN as one storage pool independent of physical location
b) Allows you to allocate more disk space on a single storage system
c) Creates LUNs that can be mapped to virtualized servers
d) Automatically stores frequently accessed data on the fastest
hard drives
12. The difference between data archiving and backup is:

a) Archiving requires more storage space than backup
b) Archives are designed to ensure integrity and retention of
fixed content, while backup creates copies of data for recovery
c) Backup provides faster recovery, while recovery from an archive can take days or weeks
d) There is no difference; data archiving is part of the backup
strategy
324
13. Fixed content is defined as:

a) Files that can be effectively compressed and deduplicated
b) Infrequently accessed data stored on inexpensive and low performance storage
c) Data that could not (or at least should not) be modified or
altered
d) Data that complies with legal regulations and requirements
14. To ensure that the stored data has not been modified, the digital
archive:
a) Periodically checks whether the stored data still matches its
cryptographic hash value
b) Performs a pair split operation, which disconnects the volume
from production severs
c) Uses password protected file access and encryption
d) Needs to be regularly audited by an external auditing company
15. The desirable value of a LUN read response time is:

a) Less than 65%
b) Less than 20ms
c) More than 100ms
d) At least 4Gb/sec
325
16. In the context of storage systems, we have several classification

categories. We can classify data according to:
a) Data source, data author and requirements on data security
b) Its importance, performance and availability requirements
c) Amount of stored data
d) Media used for data storage
17. Automated tiering is implemented because it:

a) Allows storage system scalability and upgrade
b) Ensures the best utilization of installed hard drives by automatically storing frequently used data on the most efficient
and powerful hard drives
c) Allows consolidation and centralized management of a heterogeneous IT environment
d) Monitors storage system performance
18. Enterprise level storage system implementation helps us when

we need to:
a) Ensure high performance for more than five servers
b) Use enormous disk space for video processing
c) Consolidate management of several midrange storage systems
from different vendors
d) Employ remote replication and ensure high availability
326
19. We can ensure economical operations and high utilization by:

a) Replacing old storage systems with new storage systems
equipped with large hard drives
b) Implementing virtualization both in servers and storage systems
c) Replacing all hard drives with modern solid state disks
d) Implementing modern racks and air conditioning
20. Which of the following items is NOT included in TCO calculation?

a) Insurance of the hardware
b) Hardware acquisition costs
c) Floor space
d) Software licensing costs
Correct answers:
1a, 2b, 3c, 4d, 5a, 6b, 7b, 8a, 9c, 10d, 11a, 12b, 13c, 14a, 15b, 16b, 17b,
18c, 19b, 20a
Glossary
328
Glossary
A
AaaS Archive as a Service. A cloud computing business model.
Actuator (arm) Read/write heads are

attached to a single head actuator, or
actuator arm, that moves the heads
around the platters.
AD Active Directory
Address A location of data, usually in
main memory or on a disk. A name or

token that identifies a network component. In local area networks (LANs),
for example, every node has a unique
address.
AL Arbitrated Loop. A network in which
nodes contend to send data, and only

one node at a time is able to send data.
AL-PA Arbitrated Loop Physical Address

AMS Adaptable Modular Storage
API Application Programming Interface
Application Management The processes that manage the capacity and
performance of applications
Array Group Also called a parity group,
is a group of hard disk drives (HDDs)

that form the basic unit of storage in a
subsystem. All HDDs in a parity group
must have the same physical capacity.
Array Unit A group of hard disk drives
in one RAID structure. Same as parity

group.
ASSY Assembly
Asymmetric virtualization See Outof-band virtualization.
Asynchronous An I/O operation whose

initiator does not await its completion
before proceeding with other work.
Asynchronous I/O operations enable
an initiator to have multiple concurrent I/O operations in progress. Also

called Out-of-band virtualization.
ATA Advanced Technology Attachment.
A disk drive implementation that integrates the controller on the disk drive
itself. Also known as IDE (Integrated
Drive Electronics) Advanced Technology Attachment.
Authentication The process of identifying an individual, usually based on a

username and password.
Availability Consistent direct access to

information over time
B
Back end In client/server applications,
the client part of the program is often

called the front end and the server
part is called the back end.
Backup image Data saved during an archive operation. It includes all the associated files, directories, and catalog
information of the backup operation.
Big Data Refers to data that becomes so
large in size or quantity that a dataset

becomes awkward to work with using traditional database management
systems. Big data entails data capacity
or measurement that requires terms
such as Terabyte (TB), Petabyte (PB),
Exabyte (EB), Zettabyte (ZB) or Yottabyte (YB). Note that variations of this
term are subject to proprietary trademark disputes in multiple countries at
the present time.
329
Glossary
BIOS Basic Input/Output System. A chip
located on all computer motherboards

that governs how a system boots and
operates.
BPaaS Business Process as a Service. A

cloud computing business model.
BPO Business Process Outsourcing.

Dynamic BPO services refer to the
management of partly standardized
business processes, including human
resources delivered in a pay-per-use
billing relationship or a self-service
consumption model.
Business Continuity Plan Describes

how an organization will resume partially or completely interrupted critical
functions within a predetermined time
after a disruption or a disaster. Sometimes also called a Disaster Recovery
Plan.
C
Cache Cache Memory. Intermediate
buffer between the channels and
drives. It is generally available and controlled as two areas of cache (cache
A and cache B). It may be batterybacked.
Cache hit rate When data is found in
the cache, it is called a cache hit, and

the effectiveness of a cache is judged
by its hit rate.
Cache partitioning Storage management software that allows the virtual

partitioning of cache and allocation of
it to different applications.
Capacity Capacity is the amount
of data that a storage system

or drive can store after configu-
ration and/or formatting.

Most data storage companies, including HDS, calculate capacity based on
the premise that 1KB = 1,024 bytes,
1MB = 1,024 kilobytes, 1GB = 1,024
megabytes, and 1TB = 1,024 gigabytes.
See also Terabyte (TB), Petabyte (PB),
Exabyte (EB), Zettabyte (ZB) and Yottabyte (YB).
CAPEX Capital expenditure the cost

of developing or providing non-consumable parts for the product or system. For example, the purchase of a
photocopier is the CAPEX, and the annual paper and toner cost is the OPEX.
(See OPEX).
CBI Cloud-based Integration. Provisioning of a standardized middleware platform in the cloud that can be used for
various cloud integration scenarios.
An example would be the integration

of legacy applications into the cloud
or integration of different cloud-based
applications into one application.
CCI Command Control Interface

CCIF Cloud Computing Interoperability
Forum. A standards organization active in cloud computing.
CDP Continuous Data Protection

Centralized management - Storage data
management, capacity management,

access security management, and
path management functions accomplished by software.
CF Coupling Facility
Chargeback A cloud computing term
that refers to the ability to report on
capacity and utilization by application
or dataset, charging business users or
departments based on how much they
use.
CIFS protocol Common internet file
system is a platform-independent file

sharing system. A network file system
330
Glossary
accesses protocol primarily used by

Windows clients to communicate file
access requests to Windows servers.
CLI Command Line Interface

CLPR Cache Logical Partition. Cache can
be divided into multiple virtual cache

memories to lessen I/O contention.
Cloud Computing Cloud computing
refers to applications and services that

run on a distributed network using
virtualized resources and accessed by
common Internet protocols and networking standards. It is distinguished
by the notion that resources are virtual and limitless, and that details of the
physical systems on which software
runs are abstracted from the user.
Source: Cloud Computing Bible, Barrie
Sosinsky (2011)
Cloud computing often entails an as a

service business model that may entail one or more of the following:
Archive as a Service (AaaS)
Business Process as a Service
(BPaas)
Failure as a Service (FaaS)
Infrastructure as a Service (IaaS)
IT as a Service (ITaaS)
Platform as a Service (PaaS)
Private File Tiering as a Service
(PFTaas)
Software as a Service (Saas)
SharePoint as a Service (SPaas)
SPI refers to the Software,
Platform and Infrastructure as a
Service business model.
Cloud
network
types
include
the
following:
Community cloud (or community
network cloud)
Hybrid cloud (or hybrid network
cloud)
Private cloud (or private network
cloud)
Public cloud (or public network
cloud)
Virtual private cloud (or virtual

private network cloud)
Cloud Enabler a concept, product or

solution that enables the deployment
of cloud computing. Key cloud enablers include:
Data discoverability
Data mobility
Data protection
Dynamic provisioning
Location independence
Multitenancy to ensure secure
privacy
Virtualization
Cloud Fundamental A core requirement to the deployment of cloud computing. Cloud fundamentals include:
Self service
Pay per use
Dynamic scale up and scale down
Cloud Security Alliance A standards

organization active in cloud computing.
Cluster A collection of computers that
are interconnected (typically at highspeeds) for the purpose of improving

reliability, availability, serviceability
or performance (via load balancing).
Often, clustered computers have access to a common pool of storage and
run special software to coordinate the
component computers activities.
CM Cache Memory, Cache Memory
Module. Intermediate buffer between

the channels and drives. It has a maximum of 64GB (32GB x two areas) of
capacity. It is available and controlled
as two areas of cache (cache A and
cache B). It is fully battery-backed (48
hours).
Community Network Cloud Infra-
structure shared between several organizations or groups with common

concerns.
331
Glossary
Concatenation A logical joining of two
series of data, usually represented by

the symbol |. In data communications, two or more data are often concatenated to provide a unique name
or reference (e.g., S_ID | X_ID). Volume managers concatenate disk address spaces to present a single larger
address spaces.
block size, the faster the data transfer

rate.
Data Duplication Software duplicates

data, as in remote copy or PiT snapshots. Maintains two copies of data.
Data Integrity Assurance that information will be protected from modification and corruption.
Connectivity technology A program
or devices ability to link with other

programs and devices. Connectivity
technology allows programs on a given computer to run routines or access
objects on another remote computer.
Controller A device that controls the
transfer of data from a computer to a

peripheral device (including a storage
system) and vice versa.
Controller-based
virtualization
Driven by the physical controller at
the hardware microcode level versus
at the application software layer and
integrates into the infrastructure to allow virtualization across heterogeneous storage and third party products.
Corporate governance Organizational
compliance with government-mandated regulations
CSV Comma Separated Value or Cluster

Shared Volume
Data Lifecycle Management An ap-
proach to information and storage

management. The policies, processes,
practices, services and tools used to
align the business value of data with
the most appropriate and cost-effective storage infrastructure from the
time data is created through its final
disposition. Data is aligned with business requirements through management policies and service levels associated with performance, availability,
recoverability, cost, and what ever parameters the organization defines as
critical to its operations.
Data Migration The process of moving
data from one storage device to another. In this context, data migration is
the same as Hierarchical Storage Management (HSM).
Data Pool A volume containing differential data only.
D
DAS Direct Attached Storage
Data block A fixed-size unit of data that
is transferred together. For example,

the X modem protocol transfers blocks
of 128 bytes. In general, the larger the
Data Protection Directive A major

compliance and privacy protection
initiative within the European Union
(EU) that applies to cloud computing.
Includes the Safe Harbor Agreement.
Data Striping Disk array data mapping
technique in which fixed-length sequences of virtual disk data addresses

are mapped to sequences of member
disk addresses in a regular rotating
pattern.
332
Glossary
Data Transfer Rate (DTR) The speed

at which data can be transferred.
Measured in kilobytes per second for a
CD-ROM drive, in bits per second for a
modem, and in megabytes per second
for a hard drive. Also, often called data
rate.
DNS Domain Name System

DRP Disaster Recovery Plan
DWDM Dense Wavelength Division
Multiplexing
Device Management Processes that

configure and manage storage systems
Direct Attached Storage (DAS) Storage that is directly attached to the application or file server. No other device
on the network can access the stored
data.
E
ENC Enclosure or Enclosure Controller.
es often used as the core of large

switched fabrics
The units that connect the controllers

with the Fibre Channel disks. They also
allow for online extending a system by
adding RKAs.
Disaster Recovery Plan (DRP) A plan
Ethernet A local area network (LAN)
Director class switches Larger switch-
that describes how an organization

will deal with potential disasters. It
may include the precautions taken
to either maintain or quickly resume
mission-critical functions. Sometimes
also referred to as a Business Continuity Plan.
architecture that supports clients and

servers and uses twisted pair cables
for connectivity.
Exabyte (EB) A measurement of data

or data storage. 1EB = 1,024PB.
Disk Array A linked group of one or

more physical independent hard
disk drives generally used to replace
larger, single disk drive systems. The
most common disk arrays are in
daisy chain configuration or implement RAID (Redundant Array of Independent Disks) technology.
A disk array may contain several disk
drive trays, and is structured to improve speed and increase protection
against loss of data. Disk arrays organize their data storage into Logical
Units (LUs), which appear as linear
block paces to their clients. A small
disk array, with a few disks, might support up to eight LUs; a large one, with
hundreds of disk drives, can support
thousands.
DLM Data Lifecycle Management
F
FaaS Failure as a Service. A proposed
business model for cloud computing in
which large-scale, online failure drills
are provided as a service in order to
test real cloud deployments. Concept
developed and proposed by the College of Engineering at the University
of California, Berkeley in 2011.
333
Glossary
Fabric The hardware that connects
workstations and servers to storage

devices in a SAN is referred to as a
fabric. The SAN fabric enables anyserver-to-any-storage device connectivity through the use of Fibre Channel
switching technology.
Failback The restoration of a failed sys-
tem share of a load to a replacement

component. For example, when a
failed controller in a redundant configuration is replaced, the devices that
were originally controlled by the failed
controller are usually failed back to
the replacement controller to restore
the I/O balance, and to restore failure
tolerance. Similarly, when a defective
fan or power supply is replaced, its
load, previously borne by a redundant
component, can be failed back to the
replacement part.
Failed over A mode of operation for fail-
ure tolerant systems in which a component has failed and its function has
been assumed by a redundant component. A system that protects against
single failures operating in failed over
mode is not failure tolerant, since failure of the redundant component may
render the system unable to function.
Some systems (e.g., clusters) are able
to tolerate more than one failure;
these remain failure tolerant until no
redundant component is available to
protect against further failures.
Failover A backup operation that auto-
matically switches to a standby database server or network if the primary

system fails, or is temporarily shut
down for servicing. Failover is an important fault tolerance function of
mission-critical systems that rely on
constant accessibility. Also called path
failover.
Failure tolerance The ability of a sys-
tem to continue to perform its function or at a reduced performance
level, when one or more of its components has failed. Failure tolerance in
disk subsystems is often achieved by
including redundant instances of components whose failure would make the
system inoperable, coupled with facilities that allow the redundant components to assume the function of failed
ones.
Fault Tolerant Describes a computer
system or component designed so

that, in the event of a component
failure, a backup component or procedure can immediately take its place
with no loss of service. Fault tolerance
can be provided with software, embedded in hardware, or provided by
some hybrid combination.
FC Fibre Channel or Field-Change (microcode update) or Fibre Channel.

A technology for transmitting data
between computer devices; a set of
standards for a serial I/O bus capable
of transferring data between 2 ports
FC-0 Lowest layer on fibre channel
transport. This layer represents the

physical media.
FC-1 This layer contains the 8b/10b encoding scheme.
FC-2 This layer handles framing and
protocol, frame format, sequence/exchange management and ordered set

usage.
FC-3 This layer contains common services used by multiple N_Ports in a node.
FC-4 This layer handles standards and

profiles for mapping upper level protocols like SCSI an IP onto the Fibre
Channel Protocol.
FC-AL Fibre Channel Arbitrated Loop.
A serial data transfer architecture developed by a consortium of computer

and mass storage device manufacturers, and is now being standardized by ANSI. FC-AL was designed for
334
Glossary
new mass storage devices and other

peripheral devices that require very
high bandwidth. Using optical fiber to
connect devices, FC-AL supports fullduplex data transfer rates of 100MB/
sec. FC-AL is compatible with SCSI for
high-performance storage systems.
FS File System
FTP File Transfer Protocol. A client-
server protocol that allows a user on

one computer to transfer files to and
from another computer over a TCP/IP
network.
FCIP Fibre Channel over IP, a network

storage technology that combines the
features of Fibre Channel and the Internet Protocol (IP) to connect distributed SANs over large distances. FCIP
is considered a tunneling protocol, as
it makes a transparent point-to-point
connection between geographically
separated SANs over IP networks. FCIP
relies on TCP/IP services to establish
connectivity between remote SANs
over LANs, MANs, or WANs. An advantage of FCIP is that it can use TCP/
IP as the transport while keeping Fibre
Channel fabric services intact.
Fibre Channel A serial data transfer ar-
chitecture developed by a consortium

of computer and mass storage device
manufacturers and now being standardized by ANSI. The most prominent
Fibre Channel standard is Fibre Channel Arbitrated Loop (FC-AL).
FICON Fiber Connectivity. A high-speed
input/output (I/O) interface for mainframe computer connections to storage devices. As part of IBMs S/390
server, FICON channels increase I/O
capacity through the combination of
a new architecture and faster physical
link rates to make them up to 8 times
as efficient as ESCON (Enterprise System Connection), IBMs previous fiber
optic channel standard.
Frames An ordered vector of words that

is the basic unit of data transmission in
a Fibre Channel network.
Front end In client/server applications,

the client part of the program is often
called the front end and the server
part is called the back end.
G
Gb Gigabit
GB Gigabyte
Gb/sec Gigabit per second
GB/sec Gigabyte per second
GbE Gigabit Ethernet
Gbps Gigabit per second
GBps Gigabyte per second
GBIC Gigabit Interface Converter
Global Cache Cache memory is used
on demand by multiple applications.

Use changes dynamically, as required
for READ performance between hosts/
applications/LUs.
GUI Graphical User Interface
H
HA High Availability
HBA Host Bus Adapter - An I/O adapter
that sits between the host computers

bus and the Fibre Channel loop and
manages the transfer of information
335
Glossary
between the two channels. In order to

minimize the impact on host processor performance, the host bus adapter
performs many low-level interface
functions automatically or with minimal processor involvement.
HDD Hard Disk Drive. A spindle of hard
destination address of each packet

and then forwards the packet to the
correct port. Device to which nodes on
a multi-point bus or loop are physically
connected.
Hybrid Cloud Hybrid cloud computing
refers to the combination of external

public cloud computing services and
internal resources (either a private
cloud or traditional infrastructure,
operations and applications) in a coordinated fashion to assemble a particular solution. Source: Gartner
Research.
disk platters that make up a hard drive,

which is a unit of physical storage
within a subsystem.
Head See read/write head

Heterogeneous The characteristic
of containing dissimilar elements. A

common use of this word in information technology is to describe a product as able to contain or be part of a
heterogeneous network, consisting of different manufacturers products that can interoperate.
Heterogeneous networks are made
possible by standards-conforming
hardware and software interfaces
used in common by different products,
thus allowing them to communicate
with each other. The Internet itself is
an example of a heterogeneous network.
Hybrid Network Cloud A composition
of two or more clouds (private, community or public). Each cloud remains

a unique entity but they are bound
together. A hybrid network cloud includes an interconnection.
Hypervisor Also called a virtual ma-
chine manager, a hypervisor is a hardware virtualization technique that

enables multiple operating systems
to run concurrently on the same computer. Hypervisors are often installed
on server hardware then run the guest
operating systems that act as servers.
Homogeneous Of the same or similar
Hypervisor can also refer to the interface

that is provided by Infrastructure as a
Service (IaaS) in cloud computing.
Leading hypervisors include VMware
vSphere Hypervisor (ESXi), Microsoft Hyper-V and the Xen hypervisor.
kind
Host Also called a server. Basically a cen-
tral computer that processes end-user

applications or requests.
Host LU Host Logical Unit. See also HLU.

Host Storage Domains Allows host
pooling at the LUN level and the priority access feature lets administrator
set service levels for applications.
Hub A common connection point for
devices in a network. Hubs are commonly used to connect segments of

a LAN. A hub contains multiple ports.
When a packet arrives at one port, it
is copied to the other ports so that all
segments of the LAN can see all packets. A switching hub actually reads the
I
I/O Input/Output. Term used to describe
any program, operation, or device that

transfers data to or from a computer
and to or from a peripheral device.
336
Glossary
IaaS Infrastructure as a Service. A cloud
computing business model delivering computer infrastructure, typically

a platform virtualization environment,
as a service, along with raw (block)
storage and networking. Rather than
purchasing servers, software, data
center space or network equipment,
clients buy those resources as a fully
outsourced service. Providers typically
bill such services on a utility computing basis; the amount of resources
consumed (and therefore the cost) will
typically reflect the level of activity.
Internal data bus A bus that operates

only within the internal circuitry of the
CPU, communicating among the internal caches of memory that are part of
the CPU chips design. This bus is typically rather quick and is independent
of the rest of the computers operations.
IP Internet Protocol
iSCSI Internet SCSI. Pronounced eye
skuzzy. An IP-based standard for linking data storage devices over a network and transferring data by carrying
SCSI commands over IP networks.
IDE Integrated Drive Electronics Advanced Technology. A standard designed to connect hard and removable
disk drives.
ITaaS IT as a Service. A cloud computing

business model. This general model is
an umbrella model that entails the SPI
business model (SaaS, PaaS and IaaS
Software, Platform and Infrastructure as a Service).
Index Cache Provides quick access to

indexed data on the media during a
browse/restore operation.
iFCP Internet Fibre Channel Protocol.
Allows an organization to extend Fibre

Channel storage networks over the Internet by using TCP/IP. TCP is responsible for managing congestion control
as well as error detection and recovery
services. iFCP allows an organization to
create an IP SAN fabric that minimizes
the Fibre Channel fabric component
and maximizes use of the companys
TCP/IP infrastructure.
In-band virtualization Refers to the
location of the storage network path,

between the application host servers
in the storage systems. Provides both
control and data along the same connection path. Also called symmetric
virtualization.
Interface The physical and logical ar-
rangement supporting the attachment

of any device to a connector or to another device.
Internal bus Another name for an internal data bus. Also, an expansion bus
is often referred to as an internal bus.
J
Java A widely accepted, open systems
programming language. Hitachis enterprise software products are all accessed using Java applications. This
enables storage administrators to access the Hitachi enterprise software
products from any PC or workstation
that runs a supported thin-client internet browser application and that
has TCP/IP network access to the computer on which the software product
runs.
Java VM Java Virtual Machine

JBOD Just a Bunch of Disks
337
Glossary
LAN Local Area Network. A communications network that serves clients

within a geographical area, such as a
building.
LBA Logical block address. A 28b value
that maps to a specific cylinder-headsector address on the disk.
LC Lucent connector. Fibre Channel connector that is smaller than a simplex

connector (SC).
LDEV Logical Device or Logical Device

(number). A set of physical disk partitions (all or portions of one or more
disks) that are combined so that the
subsystem sees and treats them as a
single area of data storage. Also called
a volume. An LDEV has a specific and
unique address within a subsystem.
LDEVs become LUNs to an open-systems host.
Load balancing The process of distrib-
uting processing and communications

activity evenly across a computer network so that no single device is overwhelmed. Load balancing is especially
important for networks where it is
difficult to predict the number of requests that will be issued to a server.
If one server starts to be swamped, requests are forwarded to another server with more capacity. Load balancing
can also refer to the communications
channels themselves.
LPAR Logical Partition (mode)

LU Logical Unit. Mapping number of an
LDEV.
LUN Logical Unit Number. 1 or more

LDEVs. Used only for open systems.
MAC Media Access Control. A MAC address is a unique identifier attached to

most forms of networking equipment.
MAN Metropolitan Area Network. A
communications network that generally covers a city or suburb. MAN is

very similar to a LAN except it spans
across a geographical region such as
a state. Instead of the workstations
in a LAN, the workstations in a MAN
could depict different cities in a state.
For example, the state of Texas could
have: Dallas, Austin, San Antonio. The
city could be a separate LAN and all
the cities connected together via a
switch. This topology would indicate
a MAN.
Mapping Conversion between two data
addressing spaces. For example, mapping refers to the conversion between

physical disk block addresses and the
block addresses of the virtual disks
presented to operating environments
by control software.
Mb Megabit
MB Megabyte
Metadata In database management
systems, data files are the files that

store the database information;
whereas other files, such as index files
and data dictionaries, store administrative information, known as metadata.
MIB Management Information Base.
A database of objects that can be

monitored by a network management
system. Both SNMP and RMON use
standardized MIB formats that allow
any SNMP and RMON tools to monitor
any device defined by a MIB.
338
Glossary
Microcode The lowest-level instruc-
tions that directly control a microprocessor. A single machine-language

instruction typically translates into
several microcode instructions.
Mode The state or setting of a program
or device. The term mode implies a

choice, which is that you can change
the setting and put the system in a different mode.
MP Microprocessor
MTS Multitiered Storage
Multitenancy In cloud computing, mul-
titenancy is a secure way to partition

the infrastructure (application, storage
pool and network) so multiple customers share a single resource pool. Multitenancy is one of the key ways cloud
can achieve massive economy of scale.
N
NAS Network Attached Storage. A disk
array connected to a controller that
gives access to a LAN Transport. It handles data at the file level.
NAT Network Address Translation

NDMP Network Data Management Protocol is a protocol meant to transport
data between NAS devices.
NetBIOS Network Basic Input/Output

System
Network A computer system that al-
lows sharing of resources, such as files

and peripheral hardware devices.
Network Cloud A communications

network. The word cloud by itself
may refer to any local area network
(LAN) or wide area network (WAN).

The terms computing and cloud
computing refer to services offered
on the public Internet or to a private
network that uses the same protocols
as a standard network. See also cloud
computing.
NFS protocol Network File System is a

protocol that allows a computer to access files over a network as easily as if
they were on its local disks.
Node An addressable entity connected

to an I/O bus or network, used primarily to refer to computers, storage
devices, and storage subsystems. The
component of a node that connects to
the bus or network is a port.
Node name A Name_Identifier associated with a node
O
OPEX Operational Expenditure. This is
an operating expense, operating expenditure, operational expense, or

operational expenditure, which is an
ongoing cost for running a product,
business, or system. Its counterpart is
a capital expenditure (CAPEX).
Out-of-band virtualization Refers to
systems where the controller is located outside of the SAN data path. Separates control and data on different
connection paths. Also called asymmetric virtualization.
339
Glossary
P
PaaS Platform as a Service. A cloud com-
puting business model delivering a

computing platform and solution stack
as a service. PaaS offerings facilitate
deployment of applications without
the cost and complexity of buying and
managing the underlying hardware,
software and provisioning hosting capabilities. PaaS provides all of the facilities required to support the complete
life cycle of building and delivering
web applications and services entirely
from the Internet.
PACS Picture archiving and communication systems
PAN Personal Area Network. A commu-
nications network that transmit data

wirelessly over a short distance. Bluetooth and Wi-Fi Direct are examples of
personal area networks
Parity A technique of checking wheth-
er data has been lost or written over

when it is moved from one place in
storage to another or when it is transmitted between computers.
Parity Group Also called an array

group. This is a group of hard disk
drives (HDDs) that form the basic unit
of storage in a subsystem. All HDDs
in a parity group must have the same
physical capacity.
Partitioned cache memory Separate

workloads in a storage consolidated
system by dividing cache into individually managed multiple partitions. Then
customize the partition to match the
I/O characteristics of assigned LUs
PAT Port Address Translation

PATA Parallel ATA
Path Also referred to as a transmission
channel, the path between two nodes

of a network that a data communication follows. The term can refer to
the physical cabling that connects the
nodes on a network, the signal that is
communicated over the pathway or a
sub-channel in a carrier frequency.
Path failover See Failover.

PB Petabyte
PCI Power Control Interface
PCIe Peripheral Component Interconnect Express
Performance - Speed of access or the delivery of information
Petabyte (PB) A measurement of capacity - the amount of data that a

drive or storage system can store after
formatting. 1PB = 1,024TB.
PiT Point-in-Time
PL Platter. The circular disk on which the
magnetic data is stored. Also called
motherboard or backplane.
Port In TCP/IP and UDP networks, an
endpoint to a logical connection. The

port number identifies what type of
port it is. For example, port 80 is used
for HTTP traffic.
Private Cloud A type of cloud com-
puting defined by shared capabilities

within a single company; modest economies of scale and less automation. Infrastructure and data reside inside the
companys data center behind a firewall. Comprised of licensed software
tools rather than ongoing services.
Example: An organization implements
its own virtual, scalable cloud and
business units are charged on a per
use basis.
Private Network Cloud A type of

cloud network with three characteristics: (1) Operated solely for a single or-
340
Glossary
ganization, (2) Managed internally or

by a third-party, (3) Hosted internally
or externally.
Protocol A convention or standard that
enables the communication between

two computing endpoints. In its simplest form, a protocol can be defined
as the rules governing the syntax, semantics, and synchronization of communication. Protocols may be implemented by hardware, software, or a
combination of the two. At the lowest
level, a protocol defines the behavior
of a hardware connection.
Provisioning The process of allocating

storage resources and assigning storage capacity for an application, usually
in the form of server disk drive space,
in order to optimize the performance
of a storage area network (SAN). Traditionally, this has been done by the SAN
administrator, and it can be a tedious
process. In recent years, automated
storage provisioning (also called autoprovisioning) programs have become
available. These programs can reduce
the time required for the storage provisioning process, and can free the administrator from the often distasteful
task of performing this chore manually.
Public Cloud Resources, such as applications and storage, available to the

general public over the Internet.
P-VOL Primary Volume
refers to resource reservation control

mechanisms rather than the achieved
service quality. Quality of service is the
ability to provide different priority to
different applications, users, or data
flows, or to guarantee a certain level
of performance to a data flow.
R
RAID Redundant Array of Independent
Disks, or Redundant Array of Inexpensive Disks. A group of disks that look

like a single volume to the server. RAID
improves performance by pulling a single stripe of data from multiple disks,
and improves fault-tolerance either
through mirroring or parity checking
and it is a component of a customers
SLA.
RAID-0 Striped array with no parity

RAID-1 Mirrored array and duplexing
RAID-3 Striped array with typically non-
rotating parity, optimized for long,

single-threaded transfers.
RAID-4 Striped array with typically non-
rotating parity, optimized for short,

multi-threaded transfers.
RAID-5 Striped array with typically ro-
tating parity, optimized for short, multithreaded transfers.
Q
QoS Quality of Service. In the field of
computer networking, the traffic engineering term quality of service (QoS)
RAID-6 Similar to RAID-5, but with dual
rotating parity physical disks, tolerating two physical disk failures.
RAM Random Access Memory

RAM DISK A LUN held entirely in the
cache area.
RD/WR Read/Write
Glossary
Read/Write Head - Read and write data
to the platters, typically there is one

head per platter side, and each head
is attached to a single actuator shaft.
Redundancy Backing up a component

to help ensure high availability.
Reliability (1) Level of assurance that
data will not be lost or degraded over

time. (2) An attribute of any commuter
component (software, hardware, or a
network) that consistently performs
according to its specifications.
Round robin mode A load balancing
technique which distributes data packets equally among the available paths.
Round robin DNS is usually used for
balancing the load of geographically
distributed Web servers. It works on
a rotating basis in that one server IP
address is handed out, then moves
to the back of the list; the next server
IP address is handed out, and then it
moves to the end of the list; and so
on, depending on the number of servers being used. This works in a looping
fashion.
Router A computer networking device

that forwards data packets toward
their destinations, through a process
known as routing.
RTO Recovery Time Objective. The
length of time that can be tolerated

between a disaster and recovery of
data.
341
hosted centrally in a cloud and are

typically accessed by users using a thin
client, such as a web browser via the
Internet. SaaS has become a common
delivery model for most business applications, including accounting (CRM
and ERP), invoicing (HRM), content
management (CM) and service desk
management, just to name the most
common software that runs in the
cloud. This is the fastest growing service in the cloud market today. SaaS
performs best for relatively simple
tasks in IT-constrained organizations.
SAN Storage Area Network. A network
linking computing devices to disk or

tape arrays and other devices over
Fibre Channel. It handles data at the
block level.
SAP (1) System Assist Processor (for I/O
processing), or (2) a German software

company.
SAS (1) SAN Attached Storage. Storage
elements that connect directly to a

storage area network (SAN) and provide data access services to computer
systems. (2) Serial Attached SCSI.
SATA Serial ATA. Serial Advanced Tech-
nology Attachment is a new standard

for connecting hard drives into computer systems. SATA is based on serial
signaling technology, unlike current
IDE (Integrated Drive Electronics) hard
drives that use parallel signaling.
SCSI Small Computer Systems Interface.
A parallel bus architecture and a protocol for transmitting large data blocks
up to a distance of 15m to 25m.
Server A central computer that process-
SaaS Software as a Service. A cloud
computing business model. SaaS is

a software delivery model in which
software and its associated data are
es end user applications or requests,

also called a host.
Server Virtualization The masking of
server resources, including the number and identity of individual physi-
342
Glossary
cal servers, processors, and operating

systems, from server users. The implementation of multiple isolated virtual
environments in one physical server.
Service-level Agreement SLA. A con-
tract between a network service provider and a customer that specifies,

usually in measurable terms, what
services the network service provider
will furnish. Many Internet service
providers (ISP) provide their customers with a SLA. More recently, IT departments in major enterprises have
adopted the idea of writing a service
level agreement so that services for
their customers (users in other departments within the enterprise) can
be measured, justified, and perhaps
compared with those of outsourcing
network providers.
Some metrics that SLAs may specify

include:
The percentage of the time services will be available
The number of users that can be
served simultaneously
Specific performance benchmarks
to which actual performance will
be periodically compared
The schedule for notification in
advance of network changes that
may affect users
Help desk response time for various classes of problems
Dial-in access availability
Usage statistics that will be provided
Service-Level Objective SLO. Individ-
ual performance metrics built into an

SLA. Each SLO corresponds to a single
performance characteristic relevant
to the delivery of an overall service.
Some examples of SLOs include: system availability, help desk incident resolution time, and application response
time.
SFP Small Form-Factor Pluggable mod-
ule Host connector. A specification for

a new generation of optical modular
transceivers. The devices are designed
for use with small form factor (SFF)
connectors, offer high speed and physical compactness, and are hot swappable.
SLA Service Level Agreement

SM Shared Memory or Shared Memory
Module. Stores the shared information about the subsystem and the
cache control information (director
names). This type of information is
used for the exclusive control of the
subsystem. Like CACHE, shared memory is controlled as 2 areas of memory
and fully non-volatile (sustained for
approximately seven days).
Snapshot Image A logical duplicated

volume (V VOL) of the primary volume. It is an internal volume intended
for restoration.
SNIA Storage Networking Industry As-
sociation. An association of producers

and consumers of storage networking
products, whose goal is to further storage networking technology and applications. Active in cloud computing.
SNMP Simple Network Management
Protocol. A TCP/IP protocol that was

designed for management of networks
over TCP/IP, using agents and stations.
SPaaS SharePoint as a Service. A cloud

computing business model.
Spare An object reserved for the purpose of substitution for a like object in
case of that objects failure.
SSD Solid-state Drive or Solid-State Disk

Storage pooling The ability to consolidate and manage storage resources
across storage system enclosures
where the consolidation of many appears as a single view.
343
Glossary
Striping A RAID technique for writing

a file to multiple disks on a block-byblock basis, with or without parity.
Subsystem Hardware or software that

performs a specific function within a
larger system.
Switch A fabric device providing full

bandwidth per port and high-speed
routing of data via link-level addressing.
Symmetric virtualization See In-band

virtualization.
Synchronous Operations that have a
fixed time relationship to each other.

Most commonly used to denote I/O
operations that occur in time sequence, i.e., a successor operation
does not occur until its predecessor is
complete.
Throughput The amount of data trans-
ferred from 1 place to another or processed in a specified amount of time.

Data transfer rates for disk drives and
networks are measured in terms of
throughput. Typically, throughputs are
measured in kb/sec, Mb/sec and Gb/
sec.
Tiered storage A storage strategy that
matches data classification to storage

metrics. Tiered storage is the assignment of different categories of data
to different types of storage media
in order to reduce total storage cost.
Categories may be based on levels
of protection needed, performance
requirements, frequency of use, and
other considerations. Since assigning
data to particular media may be an
ongoing and complex activity, some
vendors provide software for automatically managing the process based on a
company-defined policy.
Tiered Storage Promotion Moving
data between tiers of storage as their

availability requirements change
TLS Tape Library System
Target The system component that re-
ceives a SCSI I/O command, an open

device that operates at the request of
the initiator
TB Terabyte. 1TB = 1,024GB.

TCO Total Cost of Ownership
Topology The shape of a network or
how it is laid out. Topologies are either

physical or logical.
Track Circular segment of a hard disk or

other storage media.
Transfer Rate See Data Transfer Rate.
TCP/IP Transmission Control Protocol

over Internet Protocol
Terabyte (TB) A measurement of capacity, data or data storage. 1TB =

1,024GB.
Thin Provisioning Thin provisioning
allows storage space to be easily allocated to servers on a just-enough and

just-in-time basis.
U
UPS Uninterruptible Power Supply - A
power supply that includes a battery
to maintain power in the event of a
power outage.
344
Glossary
VI Virtual Interface. A research proto-
WAN Wide Area Network. A computing
type that is undergoing active development, and the details of the implementation may change considerably. It
is an application interface that gives user-level processes direct but protected
access to network interface cards.
This allows applications to bypass IP
processing overheads (for example,
copying data, computing checksums)
and system call overheads while still
preventing 1 process from accidentally
or maliciously tampering with or reading data being used by another.
Virtualization Referring to storage
virtualization, virtualization is the

amalgamation of multiple network
storage devices into what appears to
be a single storage unit. Storage virtualization is often used in a SAN, and
makes tasks such as archiving, backup
and recovery easier and faster. Storage
virtualization is usually implemented
via software applications.
There are many additional types of virtualization.
Virtual Private Cloud (VPC) Private

cloud existing within a shared or public cloud (for example, the Intercloud).
Also known as a virtual private network cloud.
Volume A fixed amount of storage on
a disk or tape. The term volume is often used as a synonym for the storage
medium itself, but it is possible for a
single disk to contain more than one
volume or for a volume to span more
than one disk.
V-VOL Virtual Volume
internetwork that covers a broad area

or region. Contrast with PAN, LAN and
MAN.
WORM Write Once, Read Many

WWN World Wide Name. A unique
identifier for an open-system host. It

consists of a 64b physical address (the
IEEE 48b format with a 12b extension
and a 4b prefix).
WWNN World Wide Node Name. A
globally unique 64b identifier assigned

to each Fibre Channel node process.
WWPN World Wide Port Name. A glob-
ally unique 64b identifier assigned to

each Fibre Channel port. A Fibre Channel ports WWPN is permitted to use
any of several naming authorities. Fibre Channel specifies a Network Address Authority (NAA) to distinguish
between the various name registration authorities that may be used to
identify the WWPN.
YB Yottabyte
Yottabyte A highest-end measurement
of data at the present time. 1YB =

1,024ZB, or 1 quadrillion GB. A recent
estimate (2011) is that all the computer hard drives in the world do not
contain 1YB of data.
Glossary
Z
Zettabyte (ZB) A high-end measure-
ment of data at the present time. 1ZB

= 1,024EB.
Zone A collection of Fibre Channel Ports
that are permitted to communicate

with each other via the fabric
Zoning A method of subdividing a storage area network into disjoint zones,

or subsets of nodes on the network.
Storage area network nodes outside a
zone are invisible to nodes within the
zone. Moreover, with switched SANs,
traffic within each zone may be physically isolated from traffic outside the
zone.
345
Storage Concepts
Welcome to the world of storage systems and data management!

We live in a world where information and knowledge respresent major
assets of our personal and professional lives. Nowadays, the success of
many organizations largely depends on the ability to gain knowledge
and process information. Digitization and information technology
allow us to share this knowledge with people across the globe, while
computers are able to retain information in the form of data. This book
walks you through various concepts of the technology that makes these
advancements possible. Efficient data management and powerful storage systems represent fascinating, state of the art technologies worth
exploring. Remember - data drives our lives and fuels our businesses.
This book is suitable for IT students, IT administrators (with or without
specialization in SAN environments and storage systems), IT managers
and all individuals with a thirst for expanding their IT knowledge.
We invite you to dive into this highly specialized IT field with us, learn
new and exciting information and become a highly valued professional.

Storage Concepts Storing and Managing Digital Data

Uploaded by

Copyright:

Available Formats

Storage Concepts Storing and Managing Digital Data

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Storage Concepts Storing and Managing Digital Data

Uploaded by

Copyright:

Available Formats

Storage Concepts

Storing and Managing Digital Data

Hitachi Data Systems Academy

Hitachi Data Systems Academy

HITACHI is a trademark or registered trademark of Hitachi, Ltd.

form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise,

LIMIT OF LIABILITY/DISCLAIMER OF WARRANTY: THE PUBLISHER AND THE AUTHORS MAKE NO

Produced and developed by Hitachi Data Systems Academy in cooperation

1. Brief introduction to data management and storage systems

2. Storage components and technologies

Just a Bunch of Discs (JBOD)

3. Storage networking and security

4. Business continuity and replication

5. Virtualization of storage systems

VM hardware requirements and networking

6. Archiving and file and content management

7. Storage system administration

Effective allocation of resources and reporting

8. Performance tuning, optimization and storage consolidation

Conventions used in the book

The Instructors Tip icon marks additional information that provides

The Remember icon accompanies summaries of what was discussed

Brief introduction to data

1 Brief introduction to data management and storage systems

Introduction to data management

Before we get to storage systems and their components, lets take

1 Brief introduction to data management and storage systems

1 Brief introduction to data management and storage systems

Basically, redudancy is achieved through techniques commonly called

Business sector data:

1 Brief introduction to data management and storage systems

Data that contains information about clients personal information

1 Brief introduction to data management and storage systems

State institutions data:

1 Brief introduction to data management and storage systems

is not so important. Therefore, we do not have to store this data on

Figure 1.3 Data lifecycle diagram.

1 Brief introduction to data management and storage systems

We can illustrate how the data lifecycle works by using accounting as

Structured and unstructured data

In addition to dividing the data according to purpose, its possible to

1 Brief introduction to data management and storage systems

1 Brief introduction to data management and storage systems

Data plays a crucial role in business nowadays. It is said that if a bank

New trends in data management: cloud computing

1 Brief introduction to data management and storage systems

1 Brief introduction to data management and storage systems

On the other hand, there are disadvantages the biggest problems

Data and information

Why should we distinguish the two?

1 Brief introduction to data management and storage systems

On the other hand, information is a more complex term.

1 Brief introduction to data management and storage systems

Levels of data processing

Logical and physical layer

Hard drive partitioning

1 Brief introduction to data management and storage systems

and physical levels actually work. In a model situation, we can have a

Master Boot Record

1 Brief introduction to data management and storage systems

1. Brief introduction to data management and storage systems

2. Storage components and technologies

Just a Bunch of Discs (JBOD)

3. Storage networking and security

4. Business continuity and replication

5. Virtualization of storage systems

VM hardware requirements and networking

6. Archiving and file and content management

7. Storage system administration

Effective allocation of resources and reporting

8. Performance tuning, optimization and storage consolidation