Challenges in Ubiquitous Data Management

Michael J. Franklin

EECS Computer Science Division

University of California,
Berkeley, CA 94720, USA
[email protected]

Abstract. Ubiquitous computing is a compelling vision for the future that is

moving closer to realization at an accelerating pace. The combination of global
wireless and wired connectivity along with increasingly small and powerful
devices enables a wide array of new applications that will change the nature of
computing. Beyond new devices and communications mechanisms, however,
the key technology that is required to make ubiquitous computing a reality is
data management. In this short paper, I attempt to identify and organize the
key aspects of ubiquitous computing applications and environments from a data
management perspective and outline the data management challenges that they
engender. Finally, I describe two on-going projects: Data Recharging and
Telegraph, that are addressing some of these issues.

Introduction

The confluence of ever smaller and more powerful computing and communication
devices, improved connectivity in both wired and wireless environments, and
emerging and accepted standards for data transfer and presentation (e.g., HTML,
XML, HTTP, WAP, etc.) are leading to a world in which computers are playing an
ever increasing role in people’s daily lives. It is reasonable to expect that ultimately,
such devices will be so pervasive and so critical to our activities that they will be
simply taken for granted − effectively disappearing into the background. This idea,
known as “Ubiquitous Computing”, has been a motivating vision underlying much
Computer Science research in the past decade, going back to the seminal writings of
Mark Wieser in which the term was coined [Wieser 91].
Of course, as anyone who depends upon computers or PDAs today realizes, there
is still quite some progress to be made before this vision of ubiquitous computing is
achieved. The necessary technology, however, is improving at an impressive rate and
advances furthering the realization of ubiquitous computing are announced almost
daily. Much of the research and development activity in this area is focused on
improving the devices themselves and on the technologies they use to communicate.
For devices, the emphasis has been on improving functionality, while reducing size,
cost, and power requirements. For communications, the emphasis has been on
improving bandwidth and coverage, and on developing protocols that are more
tolerant of the error and connectivity characteristics of wireless and mobile devices.

R. Wilhelm (Ed.): Informatics. 10 Years Back. 10 Years Ahead, LNCS 2000, pp. 24-33, 2001.
 Springer-Verlag Berlin Heidelberg 2001
 Challenges in Ubiquitous Data Management 25

Improved hardware and networking are clearly central to the development of

ubiquitous computing, but an equally important and difficult set of challenges revolve
around Data Management [AK93]. In order for computing to fade into the
background while supporting more and more activities, the data required to support
those activities must be reliably and efficiently stored, queried, and delivered.
Traditional approaches to data management such as caching, concurrency control,
query processing, etc. need to be adapted to the requirements and restrictions of
ubiquitous computing environments. These include resource limitations, varying and
intermittent connectivity, mobile users, and dynamic collaborations.

In this paper we first discuss the main characteristics of applications that

ubiquitous computing aims to support and then focus on the requirements that such
applications impose on data management technology. We then examine several
different aspects of data management and how they are being adapted to these new
requirements.

Applications and Data Management Requirements

While there is wide agreement on the great potential of ubiquitous computing, it is not
yet clear what the “killer applications” (i.e., the uses that will result in widespread
adoption) will be. Many researchers and product developers have created example
scenarios to demonstrate the potential of the technology. Due to the integrated and
universal nature of ubiquitous computing, these scenarios tend to include a large
number of functions rather than any one single application. Thus, some in industry
have begun to talk in terms of delivering a certain type of “user experience” rather
than a particular application or suite of applications. These scenarios tend to involve
users with several portable devices, moving between different environments (e.g.,
home, car, office, conference). The devices typically take an active (and often
annoying) role in reminding the user of various appointments and tasks that are due,
provide access to any and all information that may be relevant to these tasks, and
facilitate communication among groups of individuals involved in the tasks.

Categories of Functionality

Rather than specify yet another such scenario, it is perhaps more useful to categorize
the functionalities that such scenarios imply. This categorization can then be
examined to determine the requirements that are imposed on data management. The
functionalities can be classified into the following:

1) Support for mobility – the compactness of the devices combined with

wireless communication means that the devices can be used in mobile
situations. Thus, existing applications must be able to operate in varied
and dynamic communication and computation environments, possibly
26 Michael J. Franklin

moving from one network or service provider to another. Furthermore,

new applications that are location-centric will also be developed.

2) Context awareness – if devices become truly ubiquitous, then they will

be used constantly in a wide range of continually changing situations.
For the devices to be truly helpful, they must be aware of the
environment as well as the tasks that the user is performing or will be
performing in the near future. Context aware applications range from
intelligent notification systems that inform the user of (hopefully)
important events or data, to “smart spaces”, that is, rooms or
environments that adapt based on who is present and what they are
doing.

3) Support for collaboration – another key theme of ubiquitous computing

applications is the support of groups of people. This support consists of
communications and conferencing as well as the storage, maintenance,
delivery, and presentation of shared data. Collaborations may be
performed in real-time, if all of the participants are available, or may be
done asynchronously otherwise. In addition to supporting on-going
collaboration, access to and analysis of traces of past activities is also
required.

Adaptivity and User Interaaction

These functionalities provide a host of challenges for data management techniques,

but one requirement is present across all of them, namely, the need for adaptivity.
Mobile users and devices, changing contexts, and dynamic groups all impose
requirements for flexibility and responsiveness that are simply not addressed by most
traditional data management techniques. Thus, adaptivity is a common theme of the
techniques that we discuss in the remainder of the paper.
It is also important to note that because ubiquitous computing is intended to
augment human capabilities in the execution of various tasks, the nature of these
applications is that the user is typically interacting in real-time with the computers.
We are able to exploit this fact as part of the solution to adaptivity by, in some cases,
depending on the users to make dynamic choices or to cope with some degree of
ambiguity. A concrete example of such a design choice is the way that many
groupware systems handle concurrent access and update to shared data. Rather than
impose rules that restrict the types and degrees of interaction that users can have, as is
done by concurrency control mechanisms in traditional database systems, a
groupware data manager will typically impose less stringent rules. The relaxation of
these rules limits the extent to which the system can autonomously handle conflicts.
Thus, such systems typically handle whatever cases they can, and when they detect a
conflict that cannot be handled automatically, they simply inform the user(s) that the
conflict has occurred, and allow them to resolve it based on their knowledge of the
situation. Thus, having users “in the loop” can be leveraged to provide more adaptive
and flexible systems.
 Challenges in Ubiquitous Data Management 27

Requirements Due to Mobility

Other data management requirements are less universal across the three categories but
yet must be addressed in order to support a comprehensive ubiquitous computing
environment. For example, the issue of mobility raises a number of issues. First, the
fact that the terminals (i.e. devices) are constantly moving, and often have limited
storage capacity means that a ubiquitous computing system must be able to deliver
data to and receive data from different and changing locations. This results in the
need for various kinds of proxy solutions, where users are handed off from one proxy
to another as they move. Protocols must be constructed in such a way as to be able to
tolerate such handoffs without breaking. Mobility also raises the need for intelligent
data staging and pre-staging, so that data can be placed close to where the users will
be when they need it (particularly in slow or unreliable communications situations).
Secondly, mobility adds location as a new dimension to applications that does not
typically play a role in stationary scenarios. For example, some of the most useful
applications for mobile devices are location-centric. Consider a system that can
answer questions such as “find the drugstores within 2 miles of my current location”.
Such a system must track the location of the current user and be able to access
information based on relative locations and distances. On a broader scale, servers
may have to track large numbers of moving objects (people, cars, devices, etc.) and be
able to predict their future locations. For example, an automated traffic control
system would have to track numerous cars, including their current positions,
directions, and velocities. Location-centric computing requires special data structures
in which location information can be encoded and efficiently stored, as well as ones in
which the dynamic positions of objects can be maintained.

Requirements Due to Context-Awareness

Context-awareness imposes significant demands on the knowledge maintained by the

system and the inferencing algorithms that use that knowledge. In order to be context
aware, a system must maintain an internal representation of users’ needs, roles, and
preferences, etc. One example of such a system is a smart calendar that routes
information to a user based on knowledge of the user’s near-term schedule (as can be
determined from the user’s PIM calendar). If, for example, a user has a meeting with
a particular client scheduled for the afternoon, such a system could send the user
information that would be highly relevant to that meeting, such as data about the
client’s account, results of previous meetings with that client, news articles relevant to
the topic of the meeting, etc.
More sophisticated systems might use various types of sensors to monitor the
environment and track users’ actions so as to be able to assist in the tasks the user is
performing. Such sensor-based systems require the ability to process data streams in
real-time and to analyze and interpret such streams. Thus, data-flow processing will
play a key role in ubiquitous computing.
Whether the system obtains its context information from sensors, user input, PIM
(personal information management) applications, or some combination of these, it
28 Michael J. Franklin

must perform a good deal of processing over the data in order to be able to accurately
assess the state of the environment and the intensions of the user. Thus, context-
aware applications impose demanding requirements for inferencing and machine
learning techniques. These processes will have to cope with incomplete and
conflicting data, and will have to do so extremely efficiently in order to be able to
interact with the user in a useful and unobtrusive manner.

Requirements Due to Collaboration

The final set of requirements we discuss here are those that result from the need to
support collaborative work by dynamic and sometimes ad hoc groups of people. As
stated above, a prime requirement that stems from such applications is adaptivity. In
addition, however, there are several other types of support that are required beyond
what has already been discussed. First, there is a need for synchronization and
consistency support. At the center of any collaborative application is a set of shared
data items that can be created, accessed, modified, and deleted by participants of the
collaboration. This coordination function must be lightweight and flexible so that
many different types of interactions can be supported, ranging from unmoderated chat
facilities, to full ACID (Atomic, Consistent, Isolated, and Durable) transactions, as
provided by traditional database systems.
A second requirement stemming from collaborative applications is the need for
reliable and available storage of history. In particular, if the collaboration is to be
performed in an asynchronous manner, users must be able to access a record of what
has happened earlier in the collaboration. Likewise, if the participants in the
collaboration can change over time (e.g., due to mobility, failures, or simply due to
the nature of the collaboration), then a durable record of participants and their actions
is essential to allow new members to join and come up to speed. Such durable storage
is also useful for keeping a log of activity, that can be used later to trace through the
causes of various outcomes of the collaboration, or as input into learning and data
mining algorithms which may help optimize future collaborations.

Example Data Management Technologies – On-Going Projects

The preceding discussion addressed some of the data management challenges that
must be addressed to support ubiquitous computing scenarios and outlined the
application properties from which they arise. In this section, we briefly describe two
on going projects that are addressing some of these aspects. The first project, called
Data Recharging, exploits user interest and preference information to deliver data
updates and other relevant items to users on their portable devices. The second
project, called Telegraph, is building an adaptive data-flow processing architecture to
process long-running queries over variable streams of data, as would arise in sensor-
based and other highly dynamic data environments.
 Challenges in Ubiquitous Data Management 29

Data Recharging: Profile-Based Data Dissemination and Synchronization

Mobile devices require two key resources to function: power and data. The mobile
nature of such devices combined with limitations of size and cost makes it impractical
to keep them continually connected to the fixed power and data (i.e., the Internet)
grids. Mobile devices cope with disconnection from these grids by "caching".
Devices use rechargeable batteries for caching power, while local storage is used for
caching data. Periodically, the device-local local resources must be ``recharged'' by
connecting with the fixed power and data grids. With existing technology, however,
the process of recharging the device resident data is much more cumbersome and
error-prone than recharging the power. Power recharging can be done virtually
anywhere in the world, requires little user involvement, and works progressively −
the longer the device is recharged, the better the device-stored power becomes. In
contrast, data ``recharging'' has none of these attributes.
The Data Recharging project is developing a service and corresponding
infrastructure that permits a mobile device of any kind to plug into the Internet at any
location for any amount of time and as a result, end up with more useful data than it
had before [CFZ00]. As with power recharging, the initiation of a data charge simply
requires "plugging in" a device to the network. The longer the device is left plugged
in, the more effective the charge. Although similar to battery recharging, data
recharging is more complex; the type and amount of data delivered during a data
charge must be tailored to the needs of the user, the capabilities of the recharged
device, and the tasks that the recharged data is needed to support.
Different mobile users will have different data needs. A business traveler may
want updates of contact information, restaurant reviews and hotel pricing guides
specific to a travel destination. Students may require access to recent course notes,
required readings for the next lecture and notification about lab space as it becomes
available. Data recharging represents specifications of user needs as profiles.
Profiles can be thought of as long-running queries that continually sift through the
available data to find relevant items and determine their value to the user.
Profiles for data recharging contain three types of information: First, the profile
describes the types of data that are of interest to the user. This description must be
declarative in nature, so that it can encompass newly created data in addition to
existing data. The description must also be flexible enough to express predicates
over different types of data and media. Second, because of limitations on bandwidth,
device-local storage, and recharging time, only a bounded amount of information can
be sent to a device during data recharging. Thus, the profile must also express the
user’s preferences in terms of priorities among data items, desired resolutions of
multi-resolution items, consistency requirements, and other properties. Finally, user
context can be dynamically incorporated into the recharging process by
parameterizing the user profile with information obtained from the device-resident
Personal Information Management (PIM) applications such as the calendar, contact
list, and To Do list.
Our previous work on user profiles has focused on 1) efficiently processing
profiles over streams of XML documents (i.e., the “XFilter” system) [AF00], 2)
learning and maintaining user profiles based on explicit user feedback [CFG00], and
3) development of a large-scale, reliable system for mobile device synchronization
30 Michael J. Franklin

[DF00]. Data recharging can build upon this work, but requires the further
development of a suitable language and processing strategy for highly expressive user
profiles (i.e., that include user preference and contextual information), and the
development of a scalable, wide-area architecture that is capable of providing a data
recharging service on a global basis to potentially millions of users.

Adaptive Dataflow Query Processing

A second important aspect of ubiquitous computing environments is the variable

nature of data availability and the challenges of managing and processing dynamic
data flows. In mobile applications for example, data can move throughout the system
in order to follow the users who need it. Conversely, in mobile applications where
the data is being created at the endpoints (say, for example, a remote sensing
application) data streams into the system in an erratic fashion to be processed, stored,
and possibly acted upon by agents residing in the network. Information flows also
arise in other applications, such as data dissemination systems in which streams of
newly created and modified data must be routed to users and shared caches.
Traditional database query processing systems break down in such environments
for a number of reasons: First, they are based on static approaches to query
optimization and planning. Database systems produce query plans using simple cost
models and statistics about the data to estimate the cost of running particular plans.
In a dynamic dataflow environment, this approach simply does not work because
there are typically no reliable statistics about the data and because the arrival rates,
order, and behavior of the data streams are too unpredictable [UFA98].
Second, the exisiting approaches cannot adequately cope with failures that arise
during the processing of a query. In current database systems, if the failure of a data
source goes undetected, the query processor simply blocks, waiting for the data to
arrive. If a failure is detected, then a query is simply aborted and restarted. Neither
of these situations is appropriate in a ubiquitous computing environment in which
sources and streams behave unpredictably, and queries can be extremely long-running
(perhaps even “continuous”).
Third, existing approaches are optimized for a batch style of processing in which
the goal is to deliver an entire answer (i.e., they are optimized for the delivery of the
last result). In a ubiquitous computing environment, where users will be interacting
with the system in a fine-grained fashion, such approaches are unacceptable.
Processed data (e.g., query results, event notifications, etc.) must be passed on to the
user as soon as they are available. Furthermore, because the system is interactive, a
user may choose to modify the query on the basis of previously returned information
or other factors. Thus, the system must be able to gracefully adjust to changes in the
needs of the users [HACO+99].
The Telegraph project at UC Berkeley [HFCD+00] is investigating these issues
through the development of an adaptive dataflow processing engine. Telegraph uses a
novel approach to query execution based on “eddies”, which are dataflow control
structures that route data to query operators on an item-by-item basis [AH00].
Telegraph does not rely upon a traditional query plan, but rather, allows the ”plan” to
develop and adapt during the execution. For queries over continuous streams of data,
 Challenges in Ubiquitous Data Management 31

the system can continually adapt to changes in the data arrival rates, data
characteristics, and the availability of processing, storage, and communication
resources.
In addition to novel control structures, Telegraph also employs non-blocking,
symmetric query processing operators, such as XJoins [UF00] and Ripple Joins
[HH99], which can cope with changing and unpredictable arrival of their input data.
The challenges being addressed in the Telegraph project include the development of
cluster-based and wide-area implementations of the processing engine, the design of
fault-tolerance mechanisms (particularly for long-running queries), support for
continuous queries over sensor data and for profile-based information dissemination,
and user interface issues.

Conclusions

Ubiquitous computing is a compelling vision for the future that is moving closer to
realization at an accelerating pace. The combination of global wireless and wired
connectivity along with increasingly small and powerful devices enables a wide array
of new applications that will change the nature of computing. Beyond new devices
and communications mechanisms, however, the key technology that is required to
make ubiquitous computing a reality is data management. Data lies at the heart of all
ubiquitous computing applications, but these applications and environments impose
new and challenging requirements for data management technology.
In this short paper, I have tried to outline the key aspects of ubiquitous computing
from a data management perspective. These aspects were organized into three
categories: 1) support for mobility, 2) context-awareness, and 3) support for
collaboration. I then examined each of these to determine a set of requirements that
they impose on data management. The over-riding issue that stems from all of these
is the need for adaptivity. Thus, traditional data management techniques, which tend
to be static and fairly rigid, must be rethought in light of this emerging computing
environment.
I also described two on-going projects that are re-examining key aspects of data
management techniques: the DataRecharging project, which aims to provide data
synchronization and dissemination of highly relevant data for mobile users based on
the processing of sophisticated user profiles; and the Telegraph project, which is
developing a dynamic dataflow processing engine to efficiently and adaptively
process streams of data from a host of sources ranging from web sources to networks
of sensors.
Of course, there are a number of very important data management issues that have
not been touched upon in this short paper. First of all, the ability to interoperate
among multiple applications and types of data will depend on standards for data-
interchange, resource discovery, and inter-object communication. Great strides are
being made in these areas, and the research issues are only a small part of the effort
involved in the standardization process. Another important area, in which on-going
research plays a central role, is the development of global-scale, secure and archival
information storage utilities. An example of such a system is the OceanStore utility,
currently under development at UC Berkeley [KBCC+00].
32 Michael J. Franklin

In summary, ubiquitous computing raises a tremendous number of opportunities

and challenges for data management research and will continue to do so for the
foreseeable future. It is crucial to recognize that although it is tempting to focus on
the latest “gadget” or communications protocol, data management plays a central role
in the development of ubiquitous computing solutions and that advances in this area
will ultimately depend on our ability to solve these new and difficult data
management problems.

Acknowledgements

The ideas and opinions expressed in this article are the result of many discussions
over a number of years with colleagues in both academic and industrial settings. The
two projects discussed in this paper are on-going collaborations with several people:
The Data Recharging project is a three-university collaboration with Mitch Cherniack
of Brandeis University and Stan Zdonik of Brown University. Matt Denny and
Danny Tom of UC Berkeley have made important contributions to the project. The
Telegraph project is being done at UC Berkeley with Joe Hellerstein. Students
currently working on the project include: Sirish Chandrasekaran, Amol Deshpande,
Kris Hildrum, Nick Lanham, Sam Madden, Vijayshankar Raman, and Mehul Shah.
The work described in this paper is partially funded by DARPA as part of the
“Endeavour Expedition” under Contract Number N66001-99-2-8913, by the National
Science Foundation under the “Data Centers” project of the Information Technology
Research program, and by research funds from IBM, Siemens, Microsoft, and the
Okawa Foundation.
Thanks also to the organizers of the Dagstuhl 10th Anniversary Conference and in
particular, to Reinhard Wilhelm for his hard work and patience in putting together an
extremely successful symposium.

References

[AF00] Mehmet Altinel, Michael J. Franklin. Efficient Filtering of XML Documents for
Selective Dissemination of Information, Proceedings of the International Conference on Very
Large Data Bases, Cairo, September 2000.

[AK93] Rafael Alonso, Henry F. Korth. Database System Issues in Nomadic Computing.
Proceedings of the ACM SIGMOD Conference, Washington, D.C., June 1993, pp388-392.

[AH00] Ron Avnur, Joseph M. Hellerstein. Eddies: Continuously Adaptive Query Processing,
Procedings of the ACM SIGMOD Conference, Philadelphia, PA, June 2000,

[CFG00] Ugur Cetintemel, Michael J. Franklin, and C. Lee Giles. Self-Adaptive User Profiles
for Large-Scale Data Delivery, Proceedings of the International Conference on Data
Engineering, San Diego, CA, February, 2000, pp 622-633

[CFZ00] Mitch Cherniack, Michael J. Franklin, Stan Zdonik. Expressing User Profiles for Data
Recharging, submitted for publication, October, 2000.
 Challenges in Ubiquitous Data Management 33

[DF00] Matthew Denny, Michael J. Franklin. Edison: Database-Supported Synchronization for

PDAs, submitted for publication, November, 2000.

[HFCD+00] Joseph M. Hellerstein, Michael J. Franklin, Sirish Chandrasekaran, Amol

Deshpande, Kris Hildrum, Sam Madden, Vijayshankar Raman, Mehul Shah. Adaptive Query
Processing: Technology in Evolution, IEEE Data Engineering Bulletin, June 2000, pp 7-18.

[HACO+99] Joseph M. Hellerstein Ron Avnur, Andy Chou, Chris Olston, Vijayshankar
Raman, Tali Roth, Christian Hidber, Peter J.Haas. Interactive Data Analysis with CONTROL,
IEEE Computer 32(8), August, 1999, pp. 51-59.

[HH99] Peter J. Haas, Joseph M. Hellerstein. Ripple Joins for Online Aggregation, Procedings
of the ACM SIGMOD Conference, Philadelphia, PA, June, 1999, pp. 287-298.

[KBCC+00] John Kubiatowicz, David Bindel, Yan Chen, Steven Czerwinski, Patrick Eaton,
Dennis Geels, Ramakrishna Gummadi, Sean Rhea, Hakim Weatherspoon, Westley Weimer,
Chris Wells, and Ben Zhao. OceanStore: An Architecture for Global-Scale Persistent Storage,
Proceeedings of the Ninth Iinternational Conference on Architectural Support for
Programming Languages and
Operating Systems (ASPLOS 2000), November 2000.

[UFA98] Tolga Urhan, Michael J. Franklin, Laurent Amsaleg. Cost Based Query Scrambling
for Initial Delays,Procedings of the ACM SIGMOD Conference, Seattle, WA, June 1998, pp.
130-141.

[UF 00] Tolga Urhan, Michael J. Franklin. XJoin: A Reactively-Scheduled Pipelined Join
Operator. IEEE Data Engineering Bulletin, 23(2), June, 2000, pp. 27-33.

[Weiser 91] Weiser, Mark. The Computer for the Twenty-First Century. Scientific American.
September 1991. pp. 94-104.