Ip Multicast Explained: June 2004 Jon Hardwick Data Connection LTD

IP MULTICAST EXPLAINED
June 2004
Jon Hardwick
Data Connection Ltd.
[email protected]
Data Connection Limited

100 Church Street
Enfield, UK
Tel: +44 20 8366 1177
http://www.dataconnection.com/
Copyright © 2004 Data Connection Limited. All Rights Reserved.

TABLE OF CONTENTS
1. INTRODUCTION AND OVERVIEW ................................................................1

1.1 Introduction ...........................................................................................1
1.2 What is Multicast?.................................................................................1
1.3 History of Multicast ...............................................................................3
1.4 Overview of Multicast............................................................................4
1.5 Document Roadmap.............................................................................9
2. MULTICAST ADDRESSING .........................................................................10
2.1 Multicast Address Allocation...............................................................10
2.2 Multicast Address Scoping .................................................................11
2.3 ASM versus SSM................................................................................11
3. MULTICAST GROUP MEMBERSHIP DISCOVERY PROTOCOLS ............13
3.1 Internet Group Management Protocol (IGMP)....................................13
3.1.1 Basic IGMP operation .............................................................13
3.1.2 Sending Group Membership Queries......................................14
3.1.3 Responding to Group Membership Queries............................14
3.1.4 Improving Group Membership Latency ...................................14
3.1.5 Source Address Filtering .........................................................15
3.2 Multicast Listener Discovery (MLD) ....................................................15
3.3 IGMP/MLD Proxying ...........................................................................15
3.4 IGMP/MLD Snooping..........................................................................16
3.5 Multicast over ATM .............................................................................16
4. MULTICAST ROUTING PROTOCOLS.........................................................17
4.1 Properties of Multicast Routing Protocols...........................................17
4.1.1 Opt-in and Opt-out Protocols...................................................17
4.1.2 Source-Based and Shared Tree Protocols..............................19
4.1.3 Determining the Upstream Router...........................................22
4.1.4 Data Plane Interactions ...........................................................22
4.1.5 Summary of Multicast Routing Protocols ................................23
4.2 Protocol Independent Multicast (PIM) ................................................24
4.2.1 PIM Sparse Mode....................................................................25
4.2.2 PIM Dense Mode.....................................................................30
4.2.3 Bi-directional PIM ....................................................................33
4.2.4 RP Discovery...........................................................................35
4.2.5 Mixed-mode PIM Configurations .............................................38
4.3 Distance Vector Multicast Routing Protocol (DVMRP) .......................39
4.4 Multicast Extensions to OSPF (MOSPF)............................................39
5. INTERDOMAIN MULTICAST ROUTING ......................................................41
5.1 Multicast Source Discovery Protocol (MSDP) ....................................41
5.1.1 Alternatives to MSDP ..............................................................43
5.2 Multicast Border Routers ....................................................................44
5.3 Border Gateway Multicast Protocol (BGMP) ......................................44

6. MULTICAST SIGNALING .............................................................................46
6.1 RSVP-TE P2MP LSPs........................................................................47
6.1.1 LSP Association and Secondary Explicit Route Object (SERO) Approach
47
6.1.2 P2P LSP Merging Approach ...................................................48
6.1.3 RSVP Broadcast Approach .....................................................48
6.2 PIM P2MP LSPs .................................................................................48
6.3 Tunneling IP Multicast Traffic Through an MPLS Network.................49
6.4 Tunneling VPN IP Multicast Traffic Through a Provider Network.......54
7. MULTICAST DATA PLANE OPERATION....................................................56
7.1 Host Functionality ...............................................................................56
7.1.1 Sending Data...........................................................................56
7.1.2 Receiving Data ........................................................................56
7.2 Router Functionality............................................................................57
7.2.1 Forwarding Data ......................................................................57
7.2.2 Unicast Encapsulation and Decapsulation ..............................57
7.2.3 Packet Arrival Information .......................................................58
8. SUMMARY.....................................................................................................59
9. ABOUT DATA CONNECTION ......................................................................60
10. GLOSSARY ...................................................................................................61
11. REFERENCES...............................................................................................64
11.1 Multicast Addressing...........................................................................64
11.2 Multicast Group Membership Discovery Protocols.............................64
11.3 Multicast Routing Protocols ................................................................65
11.4 Interdomain Multicast Routing ............................................................65
11.5 Multicast Signaling..............................................................................65

1. INTRODUCTION AND OVERVIEW
1.1 Introduction
In contrast to the one-to-one model of IP unicast, in which data packets are sent from a single
source to a single recipient, IP multicast provides a method of efficient many-to-many
communication. This concept is becoming increasingly important, both in the Internet and in
private networks, for providing services such as multimedia content delivery.
This is especially true for services that require material to be propagated simultaneously to a
subset of subscribers. For example, multicast may be used in the provision of e-learning
courses, in the distribution of information to stock-market feeds or news feeds, or in
broadcasting webinars, radio and television.
There is a large and ever-expanding range of protocols used to provide and support various
aspects of multicast, including address allocation, group membership, and multicast routing and
signaling. It can be hard to keep track of and understand the protocols in all these areas.
This white paper gives an overview of the most important network-layer protocols used in
multicast. We aim to provide a guide for developers or network managers familiar with unicast
routing but new to multicast.
We begin by giving an introduction to the key concepts of multicast, starting from the ground up.
We then go into more depth about the main protocols in each area, focusing particularly on
multicast routing. Where there are multiple protocols offering similar functionality, we compare
and contrast the different approaches.
1.2 What is Multicast?
Normal IP packets are sent from a single source to a single recipient. Along the way these
packets are forwarded by a number of routers between the source and recipient, according to
forwarding table information that has been built up by configuration and routing protocol activity.
This form of IP packet delivery is known as unicast.
However, some scenarios (for example, audio/video streaming broadcasts) need individual IP
packets to be delivered to multiple destinations. Sending multiple unicast packets to achieve
this is unacceptable because
• it would require the source to hold a complete list of recipients
• multiple identical copies of the same data would flow over the same links, increasing
bandwidth requirements and costs.1
Instead, data to multiple destinations can be delivered using multicast.
1
This is particularly inefficient on transport networks (such as Ethernet) that support one-to-many
transmission.
Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 1

http://www.dataconnection.com
Multicast allows the source to send a single copy of data, using a single address for the entire
group of recipients. Routers between the source and recipients use the group address to route
the data. The routers forward duplicate data packets wherever the path to recipients diverges.
A multicast group identifies a set of recipients that are interested in a particular data stream, and
is represented by an IP address from a well-defined range, as discussed in Chapter 2,
Multicast Addressing. Data addressed to this IP address is forwarded to all members of the
multicast group.
A source host sends data to a multicast group by simply setting the destination IP address of
the datagram to be the multicast group address. Sources do not need to register in any way
before they can begin sending data to a group, and do not need to be members of the group
themselves.
In the following diagram, S2 sends a single copy of its multicast data addressed to the multicast
group. The group consists of hosts G1, G2 and G3. The data is duplicated at routers R1 and
R3 to ensure that it reaches all the hosts that are interested in this multicast data. G4 and G5
do not belong to the multicast group, and hence do not receive copies of the data.
In this and future diagrams
• S indicates a host sending multicast data

• G indicates a host which may or may not be a member of the multicast group
• R indicates a multicast-capable router.
Flow of multicast data Hosts that aren't

Network connections interested don't
S1 receive data
G4 G5
S2 sends single G2
copy of data to R9
multicast group
R2 R10
R3
R4
S2 R1
R5 R6 R7 R8
Multicast routers
make copies of data
and forward to
multiple recipients
G1
G3

Information about which parts of the network contain members of a particular multicast group is
distributed as follows.
• Hosts who wish to receive data from the multicast group join the group by sending a
message to a multicast router on a local interface, using a multicast group membership
discovery protocol such as IGMP or MLD. These protocols are discussed in Chapter 3,
Multicast Group Membership Discovery Protocols.
• Multicast routers communicate among themselves, using a multicast routing protocol such
as PIM-SM, as discussed in Chapter 4, Multicast Routing Protocols. These protocols
ensure that
• multicast traffic reaches all of the recipients that have joined the multicast group
• multicast traffic does not reach networks that do not have any such recipients (unless
the network is a transit network on the way to other recipients)
• the number of identical copies of the same data flowing over the same link is
minimized.
To satisfy these requirements, multicast routing protocols calculate a multicast distribution tree
of recipients.
1.3 History of Multicast
The need for IP multicast has been around for many years. In the early 1990s, some multicast-
enabled routers did exist, but most routers on the IP network did not have multicast capabilities.
In order to overcome this, multicast data packets were encapsulated within unicast packets and
sent down ‘tunnels’ (predetermined routes though non-multicast routers) between the multicast-
enabled routers. This network of tunnels was known as the Multicast Backbone (MBONE). On
arriving at a multicast router, the unicast packet was decapsulated into a multicast packet again.
In the late 1990s, ISPs began to replace tunnels with native multicast routing, in which
intermediate routers, although not multicast-enabled, were able to forward raw multicast packets
without encapsulating them into unicast packets first.
To date, multicast has not been deployed in networks nearly as widely as unicast. One way of
measuring the extent to which a networking technology is deployed is to count the number of
Autonomous Systems (ASs) in the Internet that have deployed it. Whereas there are over
17,000 ASs in the Internet that support unicast routing today, there are only 485 that support
multicast routing.
Support for multicast routing is, however, being slowly rolled out by service providers. The
number of ASs supporting multicast routing has gradually increased over the last three years (in
January 2004, the figure was 320). This reflects growing customer demand for multicast
services, as well as the availability of next-generation technology making multicast a viable
business case for service providers.

1.4 Overview of Multicast
Section 1.2, What is Multicast?, gave a basic introduction to the concept of IP multicast. In
this section we look in a little more detail at the way that IP multicast is carried out, and
introduce the main protocols used today. This leads into Chapters 2 to 7, where this material is
explained and discussed in depth.
Multicast Addressing
Before a host can send data to a multicast group, or join the list of recipients for a group, it must
know the address of the multicast group. Multicast IP address allocation is discussed in detail in
Chapter 2, Multicast Addressing, but in outline works as follows.
• The Internet Assigned Numbers Authority (IANA) assigns multicast group addresses for
well-known protocols and services.
• Other addresses are delegated for allocation by network administrators.
• Certain address ranges are assigned for use within local or administratively scoped
boundaries.
In general, identification of the multicast group address is an application-level issue.
When joining a multicast group, hosts can opt to receive data sent to the group from any source,
or to receive data sent to the group from one specific source only.
• To receive data from any source, hosts need to specify only the IP address of the multicast
group. This is known as Any Source Multicast (ASM).
• To receive data from one particular source only, hosts must specify both the IP address of
the multicast group and the IP address of the source. This is known as Source-Specific
Multicast (SSM).
The advantages and disadvantages of ASM and SSM are discussed in Section 2.3, ASM
versus SSM.
Multicast Group Membership Discovery Protocols
Once a host has identified a multicast group (and optionally a specific source) in which it is
interested, the next stage is for the host to register that interest with a local multicast router.
This router needs to know the multicast group memberships of all hosts directly connected to it.
Host memberships are communicated using a Multicast Group Membership Discovery (MGMD)
protocol, which runs between a router and directly connected hosts. The MGMDs used in the
Internet are IGMP for IPv4 and MLD for IPv6. Chapter 3, Multicast Group Membership
Discovery Protocols, gives more details of MGMD protocols.

Multicast Routing Protocols
When a multicast router knows the group memberships of its directly connected hosts, it
exchanges information with other routers enabling it to join or leave trees of multicast group
recipients. These exchanges use a multicast routing protocol.
The three multicast routing protocols used to any significant extent today are Protocol
Independent Multicast Sparse Mode (PIM-SM), Protocol Independent Multicast Dense Mode
(PIM-DM) and Distance Vector Multicast Routing Protocol (DVMRP), with PIM-SM being
particularly widespread. DVMRP was widely used in the MBONE in the past, but is now
recommended for use only for compatibility with existing deployments. Bi-directional PIM
(BIDIR-PIM) is still somewhat experimental and is not yet widely deployed. Two further
protocols exist, Multicast OSPF (MOSPF) and Core-Based Trees Multicast Routing (CBT), but
these have never been seriously deployed.
Once a tree has been constructed, multicast data is forwarded down the tree of recipients, with
duplicate copies of the packets generated where the tree branches. The tree of multicast group
recipients may be constructed in one of two ways, depending on the multicast routing protocol in
use.
• In opt-in protocols, routers indicate which multicast groups they want to receive data for, in
advance of that data flowing.
• In opt-out or broadcast-and-prune protocols, every router is initially assumed to want to

receive multicast data, and so each tree initially spans every link in the network. Routers
prune themselves from a given tree.
For further details, see Section 4.1.1, Opt-in and Opt-out Protocols.
The location of the root of any given multicast tree also depends on the multicast routing
protocol in use. Multicast protocols use one or both of the following methods.
• Source-based tree protocols build a separate tree for each source sending data to a
multicast group. Each tree is rooted at a router adjacent to the source, and sources send
data directly to the root of the tree.
• Shared tree protocols build a single tree used for all sources sending to a multicast group.
The tree is rooted at some selected node (in PIM, this router is called the Rendezvous
Point, or RP). The protocols then use a protocol-specific mechanism to transport the
multicast datagrams from the source to the root of the tree.
For further explanation, see Section 4.1.2, Source-Based and Shared Tree Protocols.
Another important difference between multicast routing protocols is the mechanism they provide
for routers to locate the upstream multicast router (towards the root of the multicast tree).
Routers need to communicate with the next upstream router when joining or leaving a multicast
tree. This is discussed in Section 4.1.3, Determining the Upstream Router.

An important aspect of any multicast routing protocol is the interactions with the data plane,
which by comparison with unicast routing protocols can be very complicated. In all multicast
routing protocols, information gathered in the data plane, such as measurements of bandwidth
used by each tree, must be passed to the control plane, because it influences protocol
operation. This level of interaction with the data plane is not required in unicast routing. This is
discussed briefly in Section 4.1.4, Data Plane Interactions, and in greater detail in Chapter 7,
Multicast Data Plane Operation.
A detailed description of each of the most important multicast routing protocols (PIM-SM,
PIM-DM, BIDIR-PIM, DVMRP and MOSPF) is given in Chapter 4, Multicast Routing
Protocols. As an introduction, a brief overview of each is given here.
PIM-SM
PIM-SM is an opt-in multicast routing protocol, and can use either source-based trees or shared
trees. It is the most widely used multicast routing protocol.
The diagram below shows how a multicast tree is constructed in PIM-SM. Hosts wishing to join
the group, such as G1, G2 and G3, send group membership requests (using IGMP or MLD) to
local routers. The routers then send PIM Join requests towards the root of the tree, which is R1.
PIM Join messages

Local group membership S1
requests (IGMP or MLD)
G2
R9
R10
R2 R3
G3 sends a
R4 request to
S2 R1
R8 to join
the group
R5 R7 R8
All PIM Join messages R6
flow towards R1, which
R8 forwards a PIM
is the root of the tree for
Join message
this multicast group
G1 towards R1
G3
When data is sent to the multicast group, it will flow from R1 to G1, G2 and G3, as in the
diagram shown earlier in Section 1.2, What is Multicast?.
Various methods are used by PIM-SM to identify the Rendezvous Point (root of the tree) in
shared tree networks, including the PIM Bootstrap Router (BSR) mechanism, Auto-RP,
Embedded RP and Anycast RP. These are described in Section 4.2.4, RP Discovery.

PIM-DM
PIM-DM is an opt-out multicast routing protocol that uses source-based trees. The following
diagram shows how a multicast distribution tree is constructed in PIM-DM. Multicast data is
initially sent to all hosts in the network. Routers that do not have any interested hosts, such as
R3 and R10, then send Prune messages to remove themselves from the tree.
1 Multicast data
2 PIM Prune messages S1
G2
R9
R2 R10
R3
R4
S2 R1
R5 R7 R8
R6
Data is initially forwarded to
all hosts in the network. Next,
Prune messages are sent to
remove non-interested hosts
from the tree. G1
G3
The next time data is sent to the multicast group, it does not flow to the routers that have sent
Prune messages; hence R3 and R10 do not receive multicast data.
Multicast data
S1
G2
R9
R2 R10
R3
R4
S2 R1
R5 R7 R8
R6
Data now forwarded to

only the hosts that have
not sent Prune messages G1 G3

BIDIR-PIM
BIDIR-PIM is based on PIM-SM, but allows data to flow in both directions along branches of the
multicast tree. This makes the interactions with the data plane very much simpler. BIDIR-PIM
does not support source-based trees, so SSM cannot be used with this protocol.
Typically, either PIM-SM or PIM-DM will be used throughout a multicast domain. However, they
may be used together, or in conjunction with BIDIR-PIM, within a single domain. More
information on mixed-mode PIM configurations is given in Section 4.2.5, Mixed-mode PIM
Configurations.
DVMRP
DVMRP is an opt-out protocol that uses its own distance vector algorithm to compute routes in
the network. It was widely used in the MBONE, but is now recommended for use only for
compatibility with existing deployments.
MOSPF
MOSPF is an extension of the OSPF unicast routing protocol that acts as a combined unicast
and multicast routing protocol. It differs significantly from every other multicast routing protocol
in that MOSPF neighbors do not exchange tree-building messages, such as the Join messages
used by PIM-SM or the Prune messages used by PIM-DM. Instead, the routers flood
information about the location of all receivers throughout the network, and then each MOSPF
router calculates each distribution tree on its own. MOSPF has never been seriously deployed.
For more information on the advantages, disadvantages and current deployment of each of
these multicast routing protocols, see Chapter 4, Multicast Routing Protocols.
Interdomain Multicast Routing
The protocols described above provide intradomain multicast routing, which is routing within a
single autonomous multicast routing domain. In some cases, multicasting across different
networks requires the use of additional protocols. This and other aspects of interdomain
multicasting are discussed in Chapter 5, Interdomain Multicast Routing.
Multicast Signaling
Multi-protocol label switching (MPLS) is the protocol of choice for extending IP unicast packet
forwarding to improve efficiency and to provide traffic engineering and guaranteed quality of
service.
There have been various proposals for performing IP multicast packet forwarding using MPLS,
making use of point-to-multipoint Label Switched Paths. An overview of this topic is given in
RFC 3353, but none of the detailed approaches has yet been standardized.
Chapter 6, Multicast Signaling, gives an overview of the main Internet-Drafts proposing MPLS
extensions for multicast signaling.

1.5 Document Roadmap
This section summarizes the content of the rest of this paper.
• Chapter 2, Multicast Addressing, describes various aspects of multicast IP addressing,

including how addresses are allocated, and the properties of Any Source Multicast (ASM)
versus Source Specific Multicast (SSM).
• Chapter 3, Multicast Group Membership Discovery Protocols, describes the protocols

used to manage membership of multicast groups. These include IGMP, MLD, IGMP
Proxying and IGMP Snooping, and Multicast over ATM.
• Chapter 4, Multicast Routing Protocols, describes the major multicast routing protocols
used today, including PIM-SM, PIM-DM, BIDIR-PIM, DVMRP and MOSPF. This chapter
also includes a discussion of the advantages, disadvantages and extent of the current
deployment of each major protocol.
• Chapter 5, Interdomain Multicast Routing, describes methods used for multicast routing
between different multicast domains. These include MSDP, Multicast Border Routers, and
BGMP.
• Chapter 6, Multicast Signaling, describes how the principles of MPLS are applied to
multicast IP packet forwarding situations.
• Chapter 7, Multicast Data Plane Operation, describes how the data plane forwards
multicast data, and the information that the data plane must make available to the control
plane. These interactions are crucial to the design of a multicast router.
• Chapter 8, Summary, provides a summary of the conclusions drawn by this paper on the
relative importance of the multicast protocols currently available.
• Chapter 9, About Data Connection, contains details of the author of this paper, of Data
Connection and of Data Connection’s range of portable software (including IP multicast).
• Chapter 10, Glossary, contains a glossary of some of the important terms used in this
paper.
• Chapter 11, References, provides details of references made in this paper.

2. MULTICAST ADDRESSING
Multicast groups are allocated an IP address within a well-defined range (224.0.0.0/4 for IPv4
and FF::/8 for IPv6).
This chapter discusses various aspects of multicast addressing, namely
• the division and allocation of the multicast address space
• multicast addresses with local, administrative and global scope
• the difference between Any Source Multicast (ASM) and Source Specific Multicast (SSM).
2.1 Multicast Address Allocation
The IP addresses of multicast groups are allocated very differently from unicast IP address.
• Some addresses are allocated globally by IANA for well-known services, in the same way
as TCP port numbers are allocated for unicast. In general, these are only allocated to
services that are used for network control. For example, the address 224.0.0.4 is used by
the DVMRP routing protocol. See http://www.iana.org/assignments/multicast-addresses
for a full list.
• Other addresses are delegated for allocation by network administrators. For IPv4, the
GLOP scheme (defined in RFC 3180) allocates the range 233.0.0.0/8 to the owners of
Autonomous Systems (ASs). The Unicast-Prefix-based addressing scheme (defined in
RFC 3306) provides a similar solution for IPv6 addresses. Rather than being based on an
AS number, with this system every single subnet in the Internet gets a 32-bit multicast
address range.
• Other addresses are only intended for local use, for example the administratively scoped
region 239.0.0.0/8 (see below), and so are assigned independently in each individual
network.
Source Specific Multicast (SSM), which is described in Section 2.3, ASM versus SSM, below,
has a significant effect on the address allocation issue. When SSM is used, multicast group
addresses are unique only for a particular source, and no longer have to be globally unique. To
avoid clashes, the address range 232.0.0.0/8 has been allocated for use only with SSM.

2.2 Multicast Address Scoping
Not all multicast addresses have a global scope. In particular, the following address ranges
have limited scope.
• 224.0.0.0/24 is the link-local scope region. Traffic sent to these addresses is only
transmitted over a single link. This is used for control traffic, for example that from
multicast routing protocols.
• 239.0.0.0/8 is the administratively scoped region. The key properties of administratively

scoped multicast are as follows.
• Packets addressed to administratively scoped multicast addresses do not cross

configured administrative boundaries.
• Administratively scoped multicast addresses are locally assigned, and hence are not
required to be unique across administrative boundaries.
An older form of multicast scoping was TTL scoping, which used TTL limits to restrict the flow of
multicast traffic. This form of scoping proved hard to configure and manage. Nowadays it is
recommended to use administrative scoping instead.
2.3 ASM versus SSM
The basic difference between Any Source Multicast (ASM) and Source Specific Multicast (SSM)
is as described in Section 1.4, Overview of Multicast. With SSM a host identifies a multicast
data stream with a source and group address pair (S,G), rather than by group address alone
(*,G).
SSM has the following advantages over ASM.
• Simpler address allocation. Since with SSM the multicast group address is local to the
source, no global allocation mechanism or protocol is required.
• Improved security and access control. SSM is less susceptible to denial-of-service attacks,
where unauthorized senders send to a multicast group.
• Simplified interdomain multicast routing. Receiving SSM traffic across domain boundaries
is easier than with ASM. For example, there is no need to coordinate the root of a global
shared tree, or use some other mechanism for source discovery.

However, SSM has the following drawbacks.
• Knowledge of source address. A recipient needs to know the source address of the SSM
traffic. In scenarios where there is only one source, this may be easily distributed along
with the group address. However, in situations where there are multiple, varying sources,
the problem becomes more serious and some other mechanism is necessary.
• IGMPv3. As described in Chapter 3, Multicast Group Membership Discovery

Protocols, in order for hosts to indicate their interest in particular sources, hosts and
routers require support for IGMPv3 (or MLDv2), as previous versions of these protocols do
not support this capability. IGMPv3 is becoming more widespread, but is still not
ubiquitous.
• Routing protocol support. SSM requires a multicast routing protocol that supports source-
based trees (see Section 4.1.2.1, Source-Based Tree Protocols) and source filtering.
While the most important protocols support these, BIDIR-PIM does not support source-
based trees, and MOSPF does not support source filtering.
In the future, SSM is likely to become very important, particularly for applications where there is
one source, or a small number of sources, which are unchanging. Such applications may
include audio/video broadcasting. However, despite the advantages of SSM, there are likely to
remain applications where the sources are many and varying, such as video-conferencing, for
which SSM is inappropriate.

3. MULTICAST GROUP MEMBERSHIP DISCOVERY PROTOCOLS
If a host wants to join a particular multicast group, it must inform the routers on its LAN of the
address of the multicast group it wishes to join. It is then the responsibility of a router on the
LAN to join the multicast group, as described in Chapter 4, Multicast Routing Protocols. Data
from the multicast group is sent to this router.
This chapter describes the protocols that hosts use to communicate their multicast group
memberships to neighboring routers. These are the Multicast Group Membership Discovery
protocols (MGMDs). The protocols discussed in this section include IGMP (used for IPv4
multicast groups), MLD (used for IPv6), and ATM multicast.
3.1 Internet Group Management Protocol (IGMP)
Internet Group Management Protocol (IGMP) is the MGMD protocol used for IPv4 multicast
groups. There have been three versions of IGMP. Versions 1 and 2 are very widely deployed,
version 3 less so.
The following table shows the level of support for IGMP in Microsoft Windows hosts.
IGMP Version Microsoft Windows Version

IGMPv1 Windows 95, Windows NT 4.0 (SP3 and earlier)
IGMPv2 Windows 98, Windows ME, Windows NT 4.0 (SP4 and later), Windows 2000
IGMPv3 Windows XP, Windows Server 2003
Each implementation of IGMP is required to interoperate with all previous implementations.
3.1.1 Basic IGMP operation
The following describes the basic operation of IGMP, common to all versions.
Note that a multicast router acts as both an IGMP host and an IGMP router in this and following
descriptions, and as a result can respond to its own IGMP messages.
• If a host wishes to join a new multicast group, it sends an unsolicited IGMP Report
message for that group.
• A local router picks up the IGMP Report message and uses a multicast routing protocol to
join the multicast group.
• Periodically, a special router called the Querier (see Section 3.1.2, Sending Group
Membership Queries) broadcasts IGMP Query messages onto the LAN to check which
groups the local hosts are subscribed to.

• Hosts respond to the Query messages by sending IGMP Report messages indicating their
group memberships.
• All routers on the LAN receive the Report messages and note the memberships of hosts on
the LAN. If a router does not receive a Report message for a particular group for a period
of time, the router assumes there are no more members of the group on the LAN, and
removes itself from the multicast group (but see Section 3.1.4, Improving Group
Membership Latency).
Note that all IGMP messages are raw IP datagrams, and are sent to multicast group addresses,
with a TTL of 1. Since raw IP does not provide reliable transport, some messages are sent
multiple times to aid reliability.
3.1.2 Sending Group Membership Queries
Only one router sends IGMP Query messages onto a particular LAN. This router is called the
Querier. IGMPv1 depended on the multicast routing protocol to decide which router was the
Querier. IGMPv2 introduced a Querier election process, which works as follows.
By default, a router takes the role of Querier. If a Querier receives an IGMP Query message
from a router on the same interface and with a lower IP address, it stops being the Querier. If a
router has stopped being the Querier, but does not receive an IGMP Query message within a
configured interval, it becomes the Querier again.
3.1.3 Responding to Group Membership Queries
Ordinary LAN routers typically forward multicast traffic onto all other LAN segments. Therefore,
the Querier does not need to know exactly which hosts on the LAN require data for a particular
multicast group. It only needs to know that one host requires the multicast data.
To avoid a ‘storm’ of responses to an IGMP Query message, each host that receives this
message starts a randomized timer for each group that it is a member of. When this timer pops,
the host sends an IGMP Report message, which is addressed to that group. Any other hosts
that are members of the group also receive the message, at which point they cancel their timer
for the group.
This mechanism ensures that at most one IGMP Report message is sent for each multicast
group in response to a single Query.
3.1.4 Improving Group Membership Latency
IGMPv2 introduced a Leave Group message, which is sent by a host when it leaves a multicast
group for which it was the last host to send an IGMP Report message. Receipt of this message
causes the Querier possibly to reduce the remaining lifetime of its state for the group, and to
send a group-specific IGMP Query message to the multicast group.
Note that the Leave Group message in not used with IGMPv3, as its source address filtering
mechanism (see below) provides the same functionality.

3.1.5 Source Address Filtering
IGMPv3 introduced an IGMP Version 3 Report message, to allow a host to include or exclude a
list of source addresses for each multicast group that the host is a member of. Routers merge
the source address requirements of different hosts for each group.
This feature is required to support SSM.
3.2 Multicast Listener Discovery (MLD)
Multicast Listener Discovery (MLD) is the MGMD protocol used for IPv6 multicast groups. MLD
messages are a subset of the Internet Control Message Protocol for IPv6 (ICMPv6) messages.
There are two versions of MLD. MLDv1 provides equivalent functionality to IGMPv2, and
MLDv2 provides equivalent functionality to IGMPv3. The message flows and mechanisms of
MLD are identical to those of IGMP.
3.3 IGMP/MLD Proxying
In certain network topologies, a multicast router does not actually need to run a multicast routing
protocol in order to deliver multicast traffic. Instead it can perform IGMP/MLD proxying.
One of the router’s interfaces is configured to be its ‘upstream’ interface. This is the interface to
the core network. All of its other interfaces are ‘downstream’ interfaces. The router performs
the router and host parts of IGMP/MLD on its downstream interfaces as normal, but only
performs the host part of IGMP/MLD on the upstream interface.
The router maintains a database of multicast group memberships on each of its downstream
interfaces. This has two purposes.
• The router uses a merged version of these memberships when acting as a host on the
upstream interface. It relays IGMP messages from its downstream interfaces directly to its
upstream interface, so that if one of its downstream interfaces requests a multicast group
membership, it passes the message on to the upstream interface to deal with.
• The router uses the database to forward multicast packets. When it receives a multicast
packet from its upstream interface, it forwards the packet out of all the downstream
interfaces that its database tells it have members of that multicast group.
There is a small additional requirement on multicast routing protocols in a network that uses
IGMP/MLD proxying. This is that multicast routers upstream of the IGMP/MLD proxy router(s)
must treat multicast traffic sent from within the IGMP/MLD proxying tree as if it came from a
directly connected source.
The main advantages of IGMP/MLD proxying are that such routers are cheaper, easier to
administer, and are independent of the multicast routing protocol used in the core network. The
disadvantage is a restriction on the network topology. Only a simple tree topology is supported,
in which a router has a single connection towards the core network and connections to one or
more disjoint edge networks.

3.4 IGMP/MLD Snooping
Ordinary LAN switches typically forward multicast traffic onto all other LAN segments, to ensure
that all receivers see it. However, if the receivers are sparsely distributed, this is a waste of
network resources.
IGMP/MLD snooping allows LAN switches to forward multicast traffic more intelligently (albeit by
violating the separation of functionality between the OSI communication layers). By processing
IGMP/MLD Report messages, the switch can determine the locations of interested receivers,
and suppress its forwarding of multicast traffic onto LAN segments where there are no
interested receivers.
3.5 Multicast over ATM
Some ATM networks are capable of supporting layer 3 (IP) multicast using point-to-multipoint
virtual circuits (P2MP VCs). These are set up using the Multicast Address Resolution Server
(MARS) protocol.
The MARS protocol works as follows.
• Following layer 3 requests for local interfaces to join or leave multicast groups, an ATM
endpoint registers its group memberships with the MARS for the ATM network.
• When it has a multicast packet to send, and no existing state for that destination address,
an ATM endpoint queries the MARS to determine what VC it needs to create.
There are two VCs models, which can be chosen administratively on a per-group basis.
• Full mesh – each sender creates its own P2MP VC to the receivers.
• Multicast Server (MCS) – each sender creates its own point-to-point VC to the MCS, and
the MCS creates a P2MP VC to the receivers. The MCS can be administratively chosen
on a per-group basis.
To avoid the overhead of duplicating multicast group membership discovery at layer 2 and
layer 3, MARS routers are recommended to suppress IGMP Query messages on interfaces to
networks containing a MARS, and instead query the MARS.

4. MULTICAST ROUTING PROTOCOLS
Once a multicast router knows the group memberships of its directly connected hosts, it
exchanges information with other routers to join the tree of multicast group recipients. Data sent
to the multicast group is forwarded to all branches on the tree. This section describes the
protocols used between routers to build a spanning tree of recipients for multicast data.
The three multicast routing protocols used to any significant extent today are Protocol
Independent Multicast Sparse Mode (PIM-SM), Protocol Independent Multicast Dense Mode
(PIM-DM) and Distance Vector Multicast Routing Protocol (DVMRP), with PIM-SM being
particularly widespread. DVMRP was widely used in the MBONE in the past, but is now
recommended for use only for compatibility with existing deployments. Bi-directional PIM
(BIDIR-PIM) is still somewhat experimental and is not yet widely deployed. Multicast OSPF
(MOSPF) and Core-Based Trees Multicast Routing (CBT) have never been seriously deployed.2
We first describe some general properties of multicast routing protocols. The remainder of this
section gives a brief overview of each of the most important multicast routing protocols:
PIM-SM, PIM-DM, BIDIR-PIM, DVMRP and MOSPF.
4.1 Properties of Multicast Routing Protocols
Four of the most important features of multicast routing protocols are the following.
• Whether they use opt-in or opt-out routing protocols.
• Whether they use source-based or shared trees.
• The methods they use to find the upstream router.
• The interactions they require with the data plane.
We will consider each of these features in turn. The table in Section 4.1.5, Summary of
Multicast Routing Protocols, summarizes the properties of each multicast routing protocol
with reference to the features above.
4.1.1 Opt-in and Opt-out Protocols
The basic difference between opt-in and opt-out protocols was described in Section 1.4,
Overview of Multicast. A more detailed explanation is given here.
2
Core Based Trees (CBT) is a shared tree multicast routing protocol which has never been deployed. It builds
bi-directional trees, rooted at a ‘core router’, in a similar way to BIDIR-PIM. CBT version 2 is described in RFC
2189, dating from 1997.

4.1.1.1 Opt-in Protocols
Opt-in or sparse protocols are designed on the assumption that the receivers for any particular
multicast group will be sparsely distributed throughout the network. In other words, they
assume that most subnets in the network will not want any given multicast packet. PIM-SM,
BIDIR-PIM, MOSPF and CBT are opt-in protocols.
Routers must indicate explicitly which multicast groups they want to receive data for, in advance
of that data flowing, by sending a Join message to the upstream router. By default, they are not
connected to multicast trees.
The diagram below shows routers R5, R8 and R9, which have interested hosts, sending
requests to join the multicast group rooted at R1, and the resulting flow of multicast data. Data
is only sent to routers that have opted to join the group.
1 Requests to join group

2 Resulting flow of S1 G2
multicast data
R9
R2 R3 R10
R4
S2 R1
R5 R6 R7 R8
R1 is the root of the
multicast group Requests to join the
group flow upstream
towards R1
G1 G3
4.1.1.2 Opt-out protocols
In opt-out or broadcast-and-prune or dense protocols, it is initially assumed that every router on

the network wishes to receive multicast data, and data is sent to all routers. Routers wishing to
remove themselves from the multicast tree must then send a Prune message to the upstream
router. Messages are sent when they receive a multicast datagram sent to a group or from a
source that they are not interested in. PIM-DM and DVMRP are opt-out protocols.

The following diagram shows how data is initially sent to all routers. R3 and R10, which do not
have any interested connected hosts, then send requests to leave the multicast group. The
next time data is sent down the multicast tree, it is not forwarded to these routers.
1 Initial flow of multicast data

2 Requests to leave group S1 G2
R9
R2 R10
R3
R4
S2 R1
R8
Data is initially sent to R5 R6 R7
all hosts, whether
interested or not. Some
routers then request to
leave the group. G1
G3
Opt-out protocols do not scale well in domains where most receivers do not wish to receive
data, such as the Internet, so they are mostly used for individual small domains.
4.1.2 Source-Based and Shared Tree Protocols
The location of the root of any given multicast tree depends on the multicast routing protocol in
use. Multicast routing protocols may use source-based or shared tree methods to ensure that
multicast data reaches the root of the tree in order to be forwarded downstream to all of the
recipients.
4.1.2.1 Source-Based Tree Protocols
Source-based tree protocols build a separate tree for each source that sends data to a multicast
group. Each tree is rooted at a router adjacent to the source. PIM-DM, DVMRP and MOSPF
are source-based tree protocols. In addition, PIM-SM can run in a mode where it acts as a
source-based tree protocol.
Routers wishing to join the multicast group must specify both the source and the group of the
multicast data they would like to receive, by sending an (S,G) message to the next upstream
router.

The following diagram shows a source-based tree rooted at R3. S2 sends data directly to the
root of the tree. If S1 wished to send data to the same multicast group, it would need to use a
new source-based tree based at R1.
Multicast data
S2 sends directly to S2
R3, which is the root of G2
the source-based tree R9
R2 R10
R3
R4
S1 R1
R5 R7 R8
R6
G1
G3
The advantages of source-based tree protocols are that multicast data paths are always
efficient, and they benefit from a simpler configuration than is necessary with shared tree
protocols. However, source-based tree protocols suffer from scalability problems when there
are large numbers of varying sources.
Source Specific Multicast (SSM), described in Section 2.3, ASM versus SSM, requires the use
of source-based trees. Protocols that cannot support source-based trees, such as BIDIR-PIM,
are therefore unable to use SSM.
4.1.2.2 Shared Tree Protocols
Shared tree protocols build a single tree that is used for all sources for a multicast group. The
tree is rooted at some selected node (in PIM, this router is called the Rendezvous Point, or RP).
The protocols then use a protocol-specific mechanism to transport the multicast datagrams from
the source to the root of the tree. Typically, they are encapsulated in a unicast datagram and
sent from a router adjacent to the source to the router at the root of the tree.
BIDIR-PIM and CBT are shared tree protocols. In addition, PIM-SM can run in a mode where it
acts as a shared tree protocol. Note that only opt-in multicast protocols can use shared trees.
When a LAN router wishes to join a multicast group, it does not specify the source of the group
it would like to join, but sends a (*, G) message to the next upstream router.

Shared tree protocols involve two additional factors.
• Multicast data must reach the root of the shared multicast distribution tree before it can be
forwarded on. This is achieved in one of two different ways.
• With unidirectional shared trees, each data packet is encapsulated by a source router,
sent to the root of the tree (using unicast delivery) and decapsulated.
• With bi-directional shared trees, the data flows natively back up the shared tree towards
the root.
PIM-SM uses the former approach and CBT and BIDIR-PIM use the latter, although CBT
also performs unicast encapsulation in certain circumstances.
• The root of each shared multicast tree must be selected in some manner, such as
pre-configuration or election.
The following diagram illustrates the flow of data in a unidirectional shared tree system, where
dashed arrows indicate encapsulated unicast data flowing from S2 to the root of the tree, which
is R1.
Multicast data When joining this shared tree, routers do

Unicast data S2 not need to specify a particular source
G2
R9
R2 R10
R3
R4
S1 R1
R5 R7 R8
R1 (the core router for R6
this multicast group)
receives data from S2
and converts it to
multicast data G1
G3
Shared tree protocols are better than source-based tree protocols when there are many
potential sources, but may involve inefficient data paths, as multicast packets must always be
sent to the root before being forwarded onward.
In a shared tree, the root of the tree (RP) must be selected in some manner, such as pre-
configuration or election. The methods PIM protocols use to select the RP are described in
Section 4.2.4, RP Discovery.

4.1.3 Determining the Upstream Router
All multicast routing protocols (with the exception of MOSPF) need to be able to perform a
reverse path forwarding (RPF) lookup on the unicast address of either a data source or the root
of a shared tree, in order to determine the next upstream interface for the multicast group.
The router uses the upstream interface as the outgoing interface for control packets (such as
Join and Prune messages), and as the incoming interface for multicast data. Multicast data
received on other interfaces is usually dropped or ignored, to reduce forwarding of duplicate
packets and avoid forwarding loops.
Some protocols use their own mechanism to exchange the routing information necessary to
perform RPF lookups. MOSPF uses its own link state mechanism, and DVMRP uses a distance
vector mechanism. Other protocols, namely PIM and CBT, rely on a Multicast Routing
Information Base (MRIB) populated by an external source.
The MRIB is similar to a unicast forwarding table, and may indeed be the same table used for
unicast forwarding. However, in some cases it is desirable for the MRIB and the unicast
forwarding table to differ, for example, when some routers do not support multicast. The PIM
protocol is independent of the particular unicast routing protocol used to populate the MRIB,
hence its name, Protocol Independent Multicast.
The MRIB can be configured with static routes, and can be populated by routing protocols such
as M-ISIS (Multi Topology Routing in IS-IS) and MBGP (Multiprotocol Extensions for BGP),
which distribute tagged routing information, in this case the information being tagged as being
for use by multicast routing protocols.
It is generally accepted that the principle of separating the construction of the MRIB from the
exchange of multicast routing information is a good one. Therefore PIM-DM is generally
preferred over DVMRP, except where required for backwards compatibility.
4.1.4 Data Plane Interactions
In contrast to unicast routing protocols, multicast routing protocols involve a tangled set of
interactions between the control plane and the data plane functionality of the router. A summary
of this is given below, and a more detailed discussion is presented in Chapter 7, Multicast Data
Plane Operation.
Unicast routing protocols use control messages to generate a forwarding table that the data
plane can then use to forward unicast packets. Multicast routing protocols also generate a set
of multicast forwarding table information that describes how to forward multicast packets;
however, the interactions with the data plane do not stop there.
• In some protocols, a situation may arise where multiple upstream routers are forwarding
the same traffic onto a LAN. The data plane must detect this, and inform the control plane,
to allow the control plane to elect a single forwarder and avoid the duplication of data
traffic.

• In some shared tree protocols, the data plane must be programmed with enough
information to allow it to perform the encapsulation/decapsulation described in Section
4.1.2.2, Shared Tree Protocols.
• In some protocols, the state associated to multicast groups and sources is soft state, which
is kept alive by the arrival of multicast data. This means that the data plane must inform
the control plane of the arrival of multicast packets so that it can reset its state expiry
timers.
• Some protocols, such as PIM-SM, allow transfer from a shared tree to a source-based tree.
To enable this, the data plane must inform the control plane when traffic arrives from a new
source on the shared tree, so that the control plane knows when to switch to a source-
based tree.
• In opt-out protocols, the arrival of a multicast packet may trigger a control plane Prune
message to prevent the arrival of future packets for that group on that tree.
It should be noted that while some multicast routing protocols require these interactions, others
do not. In fact BIDIR-PIM and MOSPF do not require any interactions with the data plane at all,
other than to program it with forwarding table information.
4.1.5 Summary of Multicast Routing Protocols
The following table summarizes the characteristics of each of the main multicast routing
protocols, including the version of IP with which they are compatible.
Protocol Opt-in / Supports Tree Type Upstream IP Version

Opt-out SSM Router Info Via
PIM-SM Opt-in Yes Shared or MRIB IPv4 &
source-based IPv6
PIM-DM Opt-out Yes Source-based MRIB IPv4 &
IPv6
BIDIR-PIM Opt-in No Shared MRIB IPv4 &
IPv6
DVMRP Opt-out Yes Source-based Distance vector IPv4
mechanism
MOSPF Opt-in No Source-based Link state IPv4
mechanism
The following sections describe each of the protocols listed above in more detail.

4.2 Protocol Independent Multicast (PIM)
Protocol Independent Multicast (PIM) is a collection of multicast routing protocols that share
• a common control message format
• the property of depending on unicast routing information, as described in Section 4.1.3,

Determining the Upstream Router.
PIM control messages are sent as raw IP datagrams (protocol number 103), either multicast to
the link-local ALL-PIM-ROUTERS multicast group, or unicast to a specific destination.
There are two main PIM protocols, as follows.
• PIM Sparse Mode (PIM-SM) is an opt-in protocol that uses both shared and source-based
trees. PIM-SM is the multicast routing protocol most widely used today.
• PIM Dense Mode (PIM-DM) is an opt-out protocol that uses source-based trees only. PIM-
DM is mostly used for individual small domains.
In addition there is a third PIM protocol, Bi-directional PIM (BIDIR-PIM), which is based on PIM
Sparse Mode but has significant differences. BIDIR-PIM is not yet particularly well deployed
compared to PIM-SM or PIM-DM, though it is supported in the latest Cisco and Juniper routers.
Sections 4.2.1 to 4.2.3 describe these PIM protocols, including a discussion of the advantages
and disadvantages of each protocol.
One of the important requirements of PIM-SM and BIDIR-PIM is the ability to discover the
address of the root of the group’s shared multicast distribution tree, known as its Rendezvous
Point (RP). The various methods of RP discovery are described in Section 4.2.4, RP
Discovery.
Typically, either PIM-SM or PIM-DM will be used throughout a multicast domain. However, they
may also be used together within a single domain, using Sparse Mode for some groups and
Dense Mode for others. This is known as Sparse-Dense Mode. Similarly, BIDIR-PIM may be
used on its own, or it may be used in conjunction with one or both of PIM-SM and PIM-DM.
Section 4.2.5, Mixed-mode PIM Configurations, describes how these mixed-mode
configurations work.

4.2.1 PIM Sparse Mode
There have been many implementations of PIM Sparse Mode (PIM-SM) and it is widely used
today. For a full discussion of the advantages and disadvantages of PIM-SM, see Section
4.2.1.5, PIM-SM Summary.
PIM-SM is an opt-in multicast routing protocol. To receive multicast data, routers must explicitly
inform their upstream neighbors of their interest in particular groups and sources. This is done
using PIM Join and Prune messages to join and leave a multicast distribution tree.
PIM-SM by default uses shared trees, with the trees rooted at a router called the Rendezvous
Point (RP) for a group. Data is sent from a source to the RP via encapsulation in PIM control
messages sent by unicast. PIM-SM also supports source-based trees, which may be used in
the following circumstances.
• To avoid having to encapsulate data sent to an RP, the RP may join a source-based tree.
• To optimize the data path, a last-hop router may choose to switch from the shared tree to a
source-based tree.
• For source-specific multicast (SSM), a last-hop router will join a source-based tree from the
outset.
PIM-SM is a soft-state protocol. That is, all state is timed-out a while after receiving the control
message that instantiated it. All PIM Join messages are periodically re-transmitted to keep the
state alive.
The following sections describe significant aspects of PIM-SM in more detail.
4.2.1.1 Basic shared tree forwarding procedure
In an opt-in multicast routing protocol like PIM-SM, a multicast routing tree is constructed before
any data is sent. Hosts indicate their interest in receiving data for a particular multicast group G
using a mechanism such as IGMP or MLD. One of the routers on the host’s LAN is elected the
Designated Router (DR) for the LAN. The DR is responsible for joining the multicast group and
forwarding multicast traffic to the LAN on behalf of its local receivers.

The following diagram shows how a multicast routing tree is constructed. G3 sends an
expression of interest in a particular multicast group to its DR (R8), which then sends a PIM
(*,G) Join message towards the RP for that group. G2 and G1 also send messages to their
DRs in the same way. The RP in the diagram below is R1.
PIM Join messages

Local group membership S1
information sent from G2
hosts to routers
R9
R10
R2 R3
G3 sends a
R4
R1 request to
S2
R8 to join
the group
R5 R7 R8
R6
R1 is the RP for the group.
R8 forwards the
PIM Join messages flow
Join message
towards R1
towards the RP,
G1 which is R1 G3
Routers forward the Join message by looking for the closest interface upstream towards the RP.
As the Join message is forwarded hop-by-hop towards the root of the tree, it causes (*,G)
forwarding state to be instantiated at each router it passes through.
Eventually, the Join message reaches either the RP or another router that already has (*, G)
forwarding state. As more hosts join the group G, their Join messages converge on the RP,
forming a shared distribution tree for G called the Rendezvous Point Tree (RPT).
The following diagram illustrates the flow of multicast data through the shared RPT constructed
in the previous diagram. When the source S1 sends data to the group G, the DR on the
source’s LAN (R3 in the diagram below) receives the data packets, encapsulates each one in a
PIM Register message and unicasts them to the RP (R1 in the diagram below).
When the RP receives the encapsulated data packets, it decapsulates them and forwards them
out down each branch of the RPT. Each router on the RPT receives the data from its upstream
neighbor and forwards it on, replicating the data packets as necessary. In this way the data is
sent to all interested receivers, following the opposite path to the PIM Join messages.

Native multicast data
Encapsulated data S1 G2
R9
R10
R2 R3
R4
S2 R1
Data sent from S1 is R5 R7 R8

R6
encapsulated as unicast
data at R3 and sent to R1,
where it is decapsulated
G1
G3
When a host no longer wishes to receive data for a multicast group, it signals this with IGMP,
MLD or whatever other mechanism is in use. If the host’s DR then has no dependent receivers,
it will remove itself from the RPT by sending a PIM (*,G) Prune message towards the RP.
4.2.1.2 RP switch to source-based tree
Encapsulation and decapsulation of the data packets sent from the source to the RP may be
expensive in terms of router CPU time. Therefore, the RP is permitted to join a source-based
tree, which does not require encapsulation. Instead, the root of the tree is a router adjacent to
the source.
This type of multicast tree is constructed in much the same way as above, but when sending
Join messages, routers specify both the source and the group for the multicast data they would
like to receive, sending a (S,G) message to the next upstream router.
This Join message is forwarded hop-by-hop, as with the (*,G) Join messages, causing (S,G)
forwarding state to be instantiated at each router it passes through. Eventually, the Join
message reaches either a router on the same LAN as S or another router that already has (S,G)
state, forming a branch of a source-based distribution tree called the Shortest-Path Tree (SPT).
The (S,G) data then gets forwarded down this SPT towards the RP.

The following diagram illustrates the flow of multicast data when a SPT is constructed for S1.
The RP (R1) has joined an SPT based at R3, which is on the same LAN as S1.
S1 sends data
Multicast data
directly to R3
S1 G2
Root of the
shortest-path
tree is R3 R9
R10
R2 R3
R4
S2 R1
R5 R7 R8
R6
R1 has joined a new
shortest-path tree
based at R3
G1
G3
Once the source-based tree has been joined and the data is flowing natively down the SPT to
the RP, the RP forwards data on down the RPT. It also begins discarding any received
encapsulated Register messages, since these are no longer needed after switching away from
a shared tree. To indicate that it is no longer interested in the Register messages, the RP
sends a PIM Register-Stop message to the DR. While the source is still active, the DR
periodically sends Null-Register messages to the RP (‘Null’ indicating that they contain no
encapsulated data) to indicate that the source remains active. The RP responds to each by
sending another Register-Stop message.
If another source wished to send data to the same multicast group, it would be necessary to
construct a new SPT. Source-based tree protocols generate a separate SPT for each source in
the network.
4.2.1.3 Last-hop switch to source-based tree
When data is flowing from a source to the RP (either Register-encapsulated or natively over the
SPT) and then down the RPT to receivers, it is unlikely to be following the most efficient route.
Therefore, last-hop routers are permitted to switch from the RPT to the SPT. If and when they
do so is implementation-dependent. The most common behavior is either to switch to the SPT
immediately (as soon as the first packet arrives over the RPT) or never to switch at all.
However, in principle an implementation could switch to the SPT when the data flow rate for an
(S,G) reaches a certain threshold.

To switch to the SPT, the last-hop router sends an (S,G) Join message towards the source S,
as when the RP switches to the SPT (see above). Eventually the SPT branch is set up and
data starts flowing down it. At this point, the last-hop router (or a router upstream of it) starts to
receive two copies of the (S,G) data, one from the RPT and one from the SPT. It therefore
needs to prune off the RPT, but for this source only. To do this, it sends a special Prune
message, called an (S,G,rpt) Prune, towards the RP. This overrides the (*,G) Join message
and prevents (S,G) data being forwarded down the RPT for G.
The following diagram illustrates the flow of data when last-hop router R9 joins the SPT for S1.
Last-hop router R9 has

Multicast data joined the SPT based at
S1 R3, and no longer receives G2
data from R4
R9
R10
R2 R3
R4
S2 R1
R5 R7 R8
R6
G1
G3
4.2.1.4 SSM forwarding procedure
When hosts join a multicast group and specify a specific source, PIM-SM does not use the RP
mechanism described above. Instead, the DR simply sends an (S,G) Join message towards the
source, joining the SPT immediately.

4.2.1.5 PIM-SM Summary
The advantages of PIM-SM are as follows.
• Like all PIM protocols, it is protocol-independent. In other words, its operation as a

multicast routing protocol is independent of the particular type of unicast routing protocol
operating alongside it in the network.
• It scales well across large networks.
• Sparse mode means that information only needs to be held at those routers in the network
that are part of a distribution tree.
• It supports the use of both SSM and ASM.
• It can support either shared trees (with the advantage of not needing to record per-source
state) or source-based trees (in which the data path is more efficient).
• It can use MSDP, SSM or Embedded RP for an inter-domain solution (see Chapter 5,
Interdomain Multicast Routing).
The disadvantages of PIM-SM are as follows.
• In shared trees, register-encapsulation and decapsulation between the source and RP can
be inefficient.
• Group-to-RP mapping has yet to be standardized.
• Many interactions with the data plane are required, which can affect the overall efficiency of
routers.
Version 1 of PIM-SM was created in 1995, but was never standardized by the IETF. It is now
considered obsolete, though it is still supported by Cisco and Juniper routers. Version 2 of
PIM-SM was standardized in RFC 2117 (in 1997) and updated by RFC 2362 (in 1998). Version
2 is significantly different from and incompatible with version 1. However, there were a number
of problems with RFC 2362, and a new specification of PIM-SM version 2 is currently being
produced by the IETF.
4.2.2 PIM Dense Mode
PIM Dense Mode (PIM-DM) is less common than PIM-SM, and is mostly used for individual
small domains, as it does not scale well across larger domains. For a discussion of the
advantages and disadvantages of PIM-DM, see Section 4.2.2.2, PIM-DM Summary.
PIM-DM is a multicast routing protocol designed with the opposite assumption to PIM-SM. It
assumes that the receivers for any multicast group are distributed densely (rather than sparsely)
throughout the network, and therefore that most (or at least many) subnets in the network will
want any given multicast packet. In other words, it is an opt-out protocol.

PIM-DM uses source-based trees to distribute data, as described in Section 4.1.2.1, Source-
Based Tree Protocols (shared trees are never used with opt-out protocols). However, rather
than having a mechanism in which routers explicitly join the tree, it is assumed to begin with that
every link is a branch of the tree. Links on which the data is not required are removed from the
tree using PIM Prune messages.
4.2.2.1 Forwarding operation
In a dense multicast routing protocol, when a source first starts sending data to the multicast
group, each router on the source’s LAN receives the data and forwards it to all its PIM
neighbors and to all links with directly attached receivers for the data.
Each router that receives a forwarded packet also forwards it in the same way, but only after
checking that the packet arrived on the interface closest to the source. If not, the packet is
dropped. This mechanism prevents forwarding loops from occurring.
In this way, the data is flooded to all parts of the network. The following diagram illustrates the
flooding of data from source S2 throughout a PIM-DM domain, where only hosts G1 and G2 are
interested in receiving the multicast data stream.
Multicast data Data forwarded to all hosts

S1 in the network, whether G2
interested or not
R9
R10
R2 R3
R4
S2 R1
R5 R7 R8
R6
G1
G3
There may be routers that have no need of the data, either for other PIM neighbors or for
directly connected receivers (for example, R3 in the previous diagram has no need of the data,
since it has no neighbors and no connected receivers). These routers respond to receipt of the
data by sending a PIM (S,G) Prune message upstream, which instantiate (S,G) Prune state in
the upstream router, causing it to stop forwarding the data to its downstream neighbor.

This may mean that the upstream router (R2 in the previous diagram) also no longer needs the
data, triggering it to send a Prune message to its upstream neighbor in the same way. This
‘broadcast and prune’ behavior means that eventually the data is only sent to those parts of the
network that require it. The following diagram shows the flow of Prune messages that would
occur in the situation shown in the previous diagram.
Prune messages
S1 G2
R9
R10
R2 R3
R4
S2 R1
R5 R7 R8
R6
G1
G3
The following diagram shows the resulting data flow after all the Prune messages shown above
have been processed.
Multicast data
S1 G2
R9
R10
R2 R3
R4
S2 R1
R5 R7 R8
R6
Data forwarded to only

the hosts that have not
sent Prune messages G1
G3

Eventually, the Prune state at each router will time out, and data will begin to flow back into the
parts of the network that had previously been pruned. This will trigger further Prune messages
from receivers, and the Prune state will be instantiated once more.
If a new receiver wishes to join the multicast group but is located in a part of the network that is
currently pruned from the tree, its local router sends a PIM (S,G) Graft message upstream,
instructing the upstream router to rejoin the multicast tree. Graft messages are acknowledged
with PIM Graft-Ack messages. This is the only time explicit acknowledgement is performed in
PIM.
4.2.2.2 PIM-DM Summary
The advantages of PIM-DM are as follows.
• It is an efficient protocol when the receivers genuinely are densely distributed throughout
the network.
• Like all PIM protocols, it is protocol-independent.
• It supports both SSM and ASM.
• It does not use RPs, which makes it simpler than PIM-SM to implement and deploy.
The disadvantages of PIM-DM are as follows.
• All routers need to store per-source state for every source in the domain.
• It does not scale well in domains where most receivers do not wish to receive data, such as
the Internet, so it is mostly used for individual small domains.
The development of PIM-DM has paralleled that of PIM-SM. Version 1 was created in 1995, but
was never standardized. It is now considered obsolete, though it is still supported by Cisco and
Juniper routers. Version 2 of PIM-DM is currently being standardized by the IETF. As with
PIM-SM, version 2 of PIM-DM is significantly different from and incompatible with version 1.
4.2.3 Bi-directional PIM
Bi-directional PIM (BIDIR-PIM) is a third PIM protocol based on PIM-SM. BIDIR-PIM is not yet
particularly well deployed compared to PIM-SM, PIM-DM or DVMRP, though it is supported in
the latest Cisco and Juniper routers. For a discussion of the advantages and disadvantages of
BIDIR-PIM, see Section 4.2.3.1, BIDIR-PIM Summary.
The main way BIDIR-PIM differs from PIM-SM is the method used to send data from a source to
the RP. In PIM-SM data is sent using either Register-encapsulation between the source and
the RP of a shared tree, or directly via a source-based tree. However, in BIDIR-PIM data can
flow from the source to the RP along a branch of the shared tree, which is bi-directional. This
means that data can flow in either direction along any given branch.

The following diagram illustrates the flow of data from source S3 to a group G though a
BIDIR-PIM domain, where R1 is the RP for G, and G1, G2 and G3 are subscribed to receive
multicast data from G. Data enters the tree at R7, which is on a branch of the multicast tree. It
flows both upstream towards the RP, and downstream towards receivers, duplicating itself
where required. Note that this is carried out using (*,G) ASM forwarding state.
Multicast data
S1 G2
R9
R10
R2 R3
R4
S2 R1
R5 R7 R8
R6
Multicast data enters the tree at
R7 and flows both up and down
branches of the tree, duplicating
itself as usual along the way
G1
S3 G3
BIDIR-PIM’s other major differences from PIM-SM are as follows.
• There are no source-based trees, and in fact no (S,G) state at all. There is therefore no
option for routers to switch from the shared tree onto a source-based tree, and SSM is not
supported.
• To avoid forwarding loops, for each RP one router on each LAN is elected the Designated
Forwarder (DF), which is a different method to PIM-SM. The DF is elected at RP discovery
time using a new PIM DF-Election message.
• As there is no Register-encapsulation, the PIM Register and Register-Stop messages are

never used.
• The forwarding rules are very much simpler than in PIM-SM, and there are no data-driven
events in the control plane at all.

4.2.3.1 BIDIR-PIM Summary
The advantages of BIDIR-PIM are as follows.
• The lack of (S,G) state means that it scales very well when there are many sources for
each group.
• In contrast to PIM-SM, no Register-encapsulation is required.
• It is a simple protocol to implement, with very simple interactions with the data plane.
• Like all PIM protocols, it is protocol-independent.
The disadvantages of BIDIR-PIM are as follows.
• The lack of source-based trees means that traffic is forced to remain on the possibly
inefficient shared tree.
• It only supports ASM, though in practice BIDIR-PIM would always be used in combination
with PIM-SM, with some groups (including SSM) supported on SM and some on BIDIR.
• It is not yet clear how inter-domain BIDIR-PIM is to be implemented, since it is not possible
to use MSDP, SSM or Embedded RP with BIDIR-PIM.
• Group-to-RP mapping has yet to be standardized.
There have been two proposed specifications for Bi-directional PIM. The first was described in
draft-farinacci-BIDIR-PIM, which dates from 1999. The protocol described here is a
replacement that is simpler than the first version, and has some improvements. It is described
in draft-ietf-pim-bidir.
4.2.4 RP Discovery
A PIM-SM or BIDIR-PIM router needs to be able to discover the address of the RP (the
Rendezvous Point, which is the single router at the base of the shared tree) for any given
multicast group. RP discovery protocols must be able to address the following issues.
• Load balancing / scalability. A router encounters a certain amount of load for each group
for which it is the RP, in particular for Register decapsulation in PIM-SM. Therefore, it
should ideally be possible to distribute this load evenly across several RPs.
• Fault-tolerance. An RP is a single point of failure for a multicast network, so ideally there

should be a mechanism to cope with the failure of an RP.
• Ease of configuration. Every router in a PIM-SM or BIDIR-PIM domain must have the
same set of group-to-RP mappings. Ideally, this should be achievable without requiring too
much work from the network operators.

Various methods of RP discovery are described in the following sections. These include the
following.
• Static configuration
• Bootstrap Router (BSR)
• Auto-RP
• Embedded RP
• Anycast RP
4.2.4.1 Static configuration
The simplest method from a technical point of view is to configure the group-to-RP mappings on
each router statically. The advantages of this scheme are its technical simplicity and ability to
support any desired group-to-RP mapping scheme. The disadvantages are the potentially large
amount of configuration effort required in a large network, and the lack of any support for fault
tolerance.
4.2.4.2 Bootstrap Router (BSR) Mechanism
The PIM Bootstrap Router (BSR) mechanism, which is specified in draft-ietf-PIM-SM-bsr,

works as follows.
• Certain routers in the domain are configured as Candidate BSRs. One of these routers is
elected as the BSR using PIM Bootstrap messages flooded throughout the domain.
• Certain routers in the domain are configured as Candidate RPs. Each Candidate RP
sends PIM Candidate-RP-Advertisement messages to the elected BSR.
• The BSR chooses a subset of Candidate RPs and floods this information as an RP-Set to
all other routers within the domain using PIM Bootstrap messages.
• Individual routers algorithmically determine the RP for any given group from the information
in the received RP-Set.
The BSR mechanism provides fault tolerance and load balancing, and requires little
configuration except at the Candidate BSRs and Candidate RPs. Its disadvantage is its
technical complexity.
4.2.4.3 Auto-RP
Auto-RP is a non-standardized protocol designed by Cisco that is similar to BSR, with the same
advantages and disadvantages. It works as follows.
• Certain routers in the domain are configured as Candidate RPs. Each Candidate RP
sends RP Announcement messages over UDP to the multicast group address
CISCO-RP-ANNOUNCE.

• Certain routers in the domain are configured as RP-Mapping Agents. These routers listen
to the CISCO-RP-ANNOUNCE multicast group and receive the RP Announcement
messages. They select a subset of Candidate RPs and send RP Discovery messages for
these RPs over UDP to the multicast group address CISCO-RP-DISCOVERY.
• All routers listen to the CISCO-RP-DISCOVERY group and receive the RP Discovery
messages. This information enables them to determine the RP for any given group.
In order to avoid the chicken-and-egg situation of multicast to the CISCO-RP-ANNOUNCE and

CISCO-RP-DISCOVERY groups, either these groups are configured to be in dense mode, or all
routers must have a static RP assignment for these groups.
4.2.4.4 Embedded RP
Embedded RP is a method of embedding the RP address of a group within the group address
itself. It is available only on IPv6, and uses a subset of the IPv6 address space. The encoding
of the IPv6 addresses is specified in draft-ietf-mboned-embeddedrp.
The advantage of using Embedded RP is that no configuration is required. The main

disadvantage is that the RP address becomes globally visible, making the RP an easier target
for denial-of-service attacks. Furthermore, as Embedded RP is only used for a subset of the
IPv6 multicast address space, another RP discovery mechanism is required for all other
addresses.
Embedded RP is also incompatible with BIDIR-PIM, as BIDIR-PIM relies on being able to do DF

election at RP discovery time, before data starts to arrive. This is not possible with Embedded
RP.
4.2.4.5 Anycast RP
Anycast RP (defined in RFC 3446) is not a method of RP discovery, but is a means of

bypassing the restriction of having a single RP per multicast group, giving the scalability and
fault-tolerance advantages of the more complex RP discovery systems.
Multiple RPs are configured with the same unicast IP address, and this shared address is
advertised as the RP address for the group. Sources and receivers use the nearest RP, as
determined by the IGP. In order that all the RPs know about all the sources for their groups,
MSDP (see Section 5.1, Multicast Source Discovery Protocol) must be run between the RPs,
though an extension to the PIM-SM protocol to remove the need for MSDP with Anycast RP has
been proposed (see draft-ietf-pim-anycast-rp).

The advantages of Anycast RP are two-fold.
• Load balancing / scalability. The burden of Register decapsulation is distributed across

several RPs. This has the advantage over other methods of RP load balancing that the
work is split for each group, removing the need to predict traffic patterns across the
multicast group address space.
• Fault-tolerance. If one RP fails, its sources and receivers are distributed to other RPs by
the unicast routing infrastructure.
As Anycast RP solves two of the problems Auto-RP and BSR are designed to avoid, it is usually
preferable to use Anycast RP in conjunction with a static RP configuration, rather than use Auto-
RP or BSR, as this avoids the complexities of the latter two systems while retaining their
benefits.
4.2.5 Mixed-mode PIM Configurations
Typically, a single one of PIM-SM, PIM-DM or BIDIR-PIM would be used throughout a multicast
domain. However it is possible to use a combination of the three by distributing multicast
groups between the different protocols. Each group must operate in either sparse, dense or
bi-directional mode; it is not possible to use a single group in more than one mode at once.
Because of this division, the protocols co-exist largely independently of one another.
The one way in which these protocols interact is that the same Hello protocol is used by each,
and is only run once on each LAN. The information learned from the Hello message exchange
must be shared among the three routing protocols.
The method of distributing groups between the three protocols is outside the scope of the PIM
protocols and is a matter of local configuration. Note that it is important that every router in the
domain has the same assignment of groups to protocols. The following techniques are used.
• The Bootstrap Router protocol has been extended to add a “Bi-directional” bit for each
group range. This method may be used to assign groups between sparse and bi-
directional modes if using BSR.
• Routers may be configured to use dense mode if the RP discovery mechanism (whatever
that may be) fails to find an available RP for a group, and to use sparse or bi-directional
mode otherwise.
• Router may be manually configured with group ranges for sparse, dense and bi-directional
modes.

4.3 Distance Vector Multicast Routing Protocol (DVMRP)
Distance Vector Multicast Routing Protocol (DVMRP) is generally regarded as being a suitable
protocol for small networks, particularly those running only multicast traffic (and hence lacking in
a suitable unicast routing protocol). However, it is not considered to scale suitably for larger
networks or more general applications.
The first version of DVMRP was created in 1988 and is documented in RFC 1075. It was (and
still is) widely used in the MBONE. There have since been two more versions, the second of
which is currently being standardized by the IETF. It is documented in
draft-ietf-idmr-dvmrp-v3, which is shortly to be published as a Proposed Standard RFC. It is
worth noting that due to the limited applicability of DVMRP there have been recent calls to class
the draft as “Historic” rather than “Proposed Standard”.
DVMRP is a dense multicast routing protocol that uses its own distance vector algorithm to
compute routes in the network. It is similar to PIM-DM but differs in the following significant
ways.
• It uses its own distance vector algorithm for computing routes through the network, rather
than depending on an external MRIB, as discussed in Section 4.1.3, Determining the
Upstream Router.
• The process of route exchange automatically resolves forwarding conflicts.
• Its routing algorithm has native support for tunnels between DVMRP routers through
unicast-only networks. These tunnels typically use IP-in-IP or Generic Routing
Encapsulation.
• Control messages are sent as raw IP datagrams with protocol number 2 (the same as
IGMP).
The advantages and disadvantages of DVMRP are similar to those of PIM-DM, except that
DVMRP is not independent of unicast routing protocols in the same way as PIM.
4.4 Multicast Extensions to OSPF (MOSPF)
Multicast Extensions to OSPF (MOSPF) is an extension of the unicast routing protocol OSPF
that turns it into a combined unicast and multicast routing protocol. MOSPF is documented in
RFC 1584, which dates from 1994. There has been only one version of the protocol, and it has
never been seriously deployed.
MOSPF differs significantly from every other multicast routing protocol in that there are no
protocol messages exchanged between MOSPF neighbors for building multicast distribution
trees. All the information necessary to compute the trees for every source and group is
broadcast to every router in the domain using the same mechanisms as for distributing OSPF
link-state information, but including information about multicast memberships.

MOSPF builds multicast routing trees as follows.
• MOSPF uses the same link-state mechanism as OSPF for computing and distributing
unicast routes through the domain, except that in MOSPF some links are tagged as being
multicast-capable and some are not. This allows incongruent network topologies for
unicast and multicast data.
• The same mechanism is also used to distribute local membership information to all routers
in the domain. This state is (*,G) only – there is no support for SSM.
• On receipt of an (S,G) multicast data packet, a router uses the route and local membership
information it has learned to compute the whole source-based tree for that (S,G). If the
router finds that it is on the tree, it forwards the data packet to just those links that require
it.
Because every router knows the same information, they all compute the same tree, and so no
Assert, Designated Router or Designated Forwarder mechanisms are necessary.
An advantage of MOSPF is that it is not necessary to define RPs. MOSPF may also have a
particular appeal in networks that already use OSPF as their unicast routing protocol, although it
has the same inherent scalability limitations as OSPF. Another disadvantage of MOSPF is that
it does not support SSM.
MOSPF is documented in RFC 1584, which dates from 1994.

5. INTERDOMAIN MULTICAST ROUTING
The multicast routing protocols described in the previous chapter apply to multicast routing
within a single domain. This chapter describes various methods of interdomain multicast
routing.
We first describe the Multicast Source Discovery Protocol (MSDP), which is used to exchange
source information between PIM-SM RPs in different domains. In current networks, multicast is
mainly deployed with domains running PIM-SM, connected using MSDP. This is not a long-
term scalable solution, as the amount of traffic flowing over MSDP scales linearly with the
number of sources in the whole Internet. We discuss potential future alternatives to MSDP.
Next, we discuss the rules for the operation of multicast border routers, which are routers that
belong to more than one multicast domain. Finally, we describe the Border Gateway Multicast
Protocol (BGMP), which is intended to provide true interdomain multicast routing between
domains running different multicast routing protocols.
5.1 Multicast Source Discovery Protocol (MSDP)
PIM-SM (described in Section 4.2.1, PIM Sparse Mode) enables multicast packets from
sources in a domain to reach receivers in the same domain, using an RP inside that domain.
The Multicast Source Discovery Protocol (MSDP) is a companion protocol that allows PIM-SM
RPs in different domains to exchange information about the multicast sources in their domains.
This information is essentially a list of source/group pairs for the active multicast sources in their
domain, which allows other PIM-SM RPs to discover and join to sources in the domain. MSDP
is described in RFC 3618 and in draft-ietf-mboned-msdp-deploy.
MSDP can also be used to exchange multicast routing information between Anycast RPs in the
same domain (see Section 4.2.4.5, Anycast RP).
MSDP works as follows.
• The RPs in each domain run MSDP. It is also possible for non-RPs to run MSDP; they can
relay MSDP information to other MSDP nodes, but do not originate or act on it.
• Each MSDP node is explicitly configured with a set of MSDP peers. MSDP peers set up
TCP connections between themselves.
• When an RP receives a PIM Register message from a new source within its domain, it
creates an MSDP Source-Active (SA) message, containing the address of the source, the
destination group and the originating MSDP node. It periodically resends this SA
MESSAGE for as long as the source is active.

• MSDP messages, containing one or more SA messages, are flooded throughout the
MSDP network. To reduce unnecessary forwarding of MSDP information, this flooding
mechanism usually relies on the BGP topology and on BGP-distributed information about
the route to the originating MSDP node.
• When an RP receives a new SA message, and one or more receivers in its domain want to
receive the multicast traffic for that source/group pair, it creates a shortest-path tree to the
source by sending a PIM Join message to the source, exactly as if it had received a PIM
Register from the source.
MSDP nodes are permitted to filter the SA messages that they originate, and for administratively
scoped groups (as described in Section 2.2, Multicast Address Scoping), the SA messages
that they forward.
Using MSDP with protocols other than PIM-SM is permitted, but there are currently no
specifications that do this.
In the following example, G1 has already sent a PIM Join message to R3, instructing it to join
the multicast group.
1. In domain A, source S1 begins to send multicast traffic to R1.
2. R1 creates an MSDP SA message, and floods it throughout the MSDP network, sending it
to R2 and R3.
3. R3 knows that it has a receiver G1 that is interested in this multicast group, so it sends a
PIM Join to S. R2 does not have any interested receivers, so it does not send a PIM Join
message to S.
Link Domain B
R2
MSDP adjacency
2 2
Domain A Domain C
R1 1 3 R3
3
3 R
S1
S G1

R3 has now indicated that it wishes to receive multicast data, and R2 has indicated it does not.
4. S1 sends multicast traffic to R1 and R3.3
5. R1 periodically continues to flood the MSDP SA MESSAGE throughout the MSDP network.
Link Domain B
R2
MSDP adjacency
5 5
Domain A Domain C
R1 4 4 R3
4 4
S 4 R
G1
S1
5.1.1 Alternatives to MSDP
MSDP is problematic as the amount of traffic flowing over MSDP scales linearly with the number
of sources in the whole Internet, meaning that it is not a long-term scalable solution. There are
two alternatives to the MSDP protocol.
• SSM (described in Section 2.3, ASM versus SSM): MSDP is not needed for SSM, as no
source discovery is necessary.
• Embedded RP (described in Section 4.2.4.4, Embedded RP): MSDP is not needed with
Embedded RP, as for an Embedded RP group there is only a single RP in the whole
Internet (rather than an RP per domain with normal groups). Embedded RP is only
supported for IPv6.
3
In this example, there are no receivers in Domain A, so R1 would actually have sent a PIM Register-Stop to
S1, causing it to stop sending the multicast traffic to R1. S1 would still send periodic PIM Null-Register
messages to R1, causing it to continue flooding the MSDP SA message. Meanwhile, PIM-SM could choose to
optimize the multicast data flow between S1 and G1 by switching G1 onto the SPT.

In practice, both SSM and Embedded RP are relatively new. There are applications for which
SSM is not ideally suited, and disadvantages to Embedded RP, notably that it is only supported
for IPv6.
Interdomain PIM-SM will require MSDP for the foreseeable future, but SSM and Embedded RP
are likely to become more important with time.
5.2 Multicast Border Routers
RFC 2715 specifies the behavior required at multicast border routers to allow domains running
different multicast routing protocols to interoperate. In order for these rules to apply, there must
either be a domain hierarchy (a tree with one domain as root), or an interdomain routing
protocol (such as BGMP, which is described in Section 5.3, Border Gateway Multicast
Protocol) must be used.
The important features of the rules for multicast border routers may be summarized as follows.
• The router consists of two or more multicast routing components, each owning a subset of
the router’s interfaces over which it runs some multicast routing protocol.
• No interface runs more than one multicast routing protocol. Hence each multicast-enabled
interface is owned by exactly one multicast routing component.
• The components communicate to tell each other which multicast data is required in each
domain.
• All the components share a common forwarding cache.
• The component that owns an interface chooses whether to accept or reject packets
received on that interface.
• Once an incoming packet has been accepted, the component that owns an interface
chooses whether or not the packet should be forwarded out of that interface.
5.3 Border Gateway Multicast Protocol (BGMP)
The goal of the Border Gateway Multicast Protocol (BGMP) is to allow full interdomain multicast
routing in a network with domains running different multicast routing protocols. BGMP is not
currently deployable, and is not likely to be in the near future. Eventually, however, BGMP or a
similar protocol may be the answer for interdomain multicast.
BGMP supports both Source-Specific Multicast (SSM) and Any Source Multicast (ASM) models,
and allows receivers to build bi-directional shared trees that span domains. Within each
domain, the multicast routing protocol routes the multicast traffic. For a given tree of domains, a
single domain is designated (by configuration) the ‘root domain’ for the tree. This is analogous
to an RP being the root of a tree of routers.

The ASM model requires that each multicast group is associated with a single root domain.
This association is encoded in IPv6 multicast group addresses using Unicast-Prefix-based
addressing. Alternatively, another mechanism, such as the Multicast Address-Set Claim
(MASC) protocol, can be used to distribute the association.
BGMP is still evolving, but its key features are as follows.
• BGMP runs on the border routers between multicast-capable domains.
• BGMP creates TCP connections between peers.
• BGMP routers send Join and Prune messages to peers in other domains.
• A BGMP router uses these Join and Prune messages to build up state for forwarding
packets between domains.
• If a BGMP router does not have forwarding state for a multicast data packet, it forwards it
towards the root domain.
BGMP is described in draft-ietf-bgmp-spec, which is soon to be published as an Informational

RFC. The status of “Informational” was chosen to underline the fact that the protocol is not
currently deployable, and that its future as an IETF standard is uncertain.

6. MULTICAST SIGNALING
Signaling protocols are used to enhance the efficiency of unicast IP packet forwarding, and to
provide traffic engineering and guaranteed quality of service.
Multi Protocol Label Switching (MPLS) is a collection of such signaling protocols. The protocols
specify point-to-point Label Switched Paths (P2P LSPs) which traverse routers along a given
data path. IP traffic is then tunneled through these LSPs, providing two key advantages.
• When a data packet enters the LSP, it is assigned a label. Transit routers along that LSP
then forward the data based on a quick look-up on that label. This is more efficient than,
say, an IP address best-match.
• When the RSVP-TE protocol is used to set up the LSP, the path of the LSP is chosen to
provide a specific quality of service. Each router along that path can set aside resources
for use solely by that LSP.
LSP tunneling is also an efficient and secure way to implement standards-based Virtual Private
Networks (VPNs), because the forwarding of data packets is based solely on the MPLS label,
so inspection of the packet contents (including the destination address) is not required. More
information on these issues is given in Data Connection’s White Paper, VPN Technologies – A
Comparison, which is available at http://www.dataconnection.com/products/whitepapers.htm.
The principles described above can also be applied to multicast IP packet forwarding situations
by creating point-to-multipoint (P2MP) LSPs. There are a number of protocol drafts that target
this area. This section gives an overview of these approaches, which include
• RSVP-TE P2MP LSPs
• PIM P2MP LSPs
• tunneling IP multicast traffic through an MPLS network
• tunneling VPN IP multicast traffic through a provider network.

6.1 RSVP-TE P2MP LSPs
RSVP-TE is the only standardized MPLS signaling protocol which provides traffic engineering,
and hence is the obvious choice to adapt to multicast signaling for applications (such as
streaming video broadcast) which have strict requirements on reliability. The requirements for
RSVP-TE extensions for P2MP are set out in draft-ietf-mpls-p2mp-requirement.
All of the suggested solutions require the ingress node to calculate the entire P2MP LSP, so
that it must somehow determine the set of receivers. This is in contrast to plain IP multicast,
where the source/RP knows only the next hop. This requirement affects the routing protocol.
See Chapter 4, Multicast Routing Protocols, for more details.
However, this requirement allows far more flexibility in choosing the tree. Rather than simply
using a shortest-path tree, the path could be optimized to, for example, have the least overall
cost (a Steiner tree).
6.1.1 LSP Association and Secondary Explicit Route Object (SERO) Approach
draft-yasukawa-mpls-rsvp-p2mp defines a method of grouping together a set of P2P LSPs to

make a P2MP LSP such that there is no duplication of resources either in the control plane or
the data plane.
This is achieved by the ingress (source) node first discovering the set of egress (receiver) nodes
and calculating the tree route for the P2MP. It then sends a single Path message with a single
Explicit Route Object (ERO) corresponding to one of the egresses, and multiple Secondary
Explicit Route Objects (SEROs) corresponding to the other egresses. The first hop in each
SERO is a branch node. Each branch node takes its corresponding SERO, and sets up a new
P2P LSP (in a different session) with an ERO taken from that SERO. The ASSOCIATION
object is used to associate all these P2P LSPs together as a P2MP LSP.
• The key advantage of this approach is that it does not duplicate state in the control plane.
The P2P LSPs making up the P2MP LSP do not overlap.
• The main disadvantage is that it does not support all of the normal RSVP-TE extensions
(e.g. Fast Reroute and other protection/restoration mechanisms), although it does support
make-before-break of the entire tree.
• A complete set of SEROs for a P2MP LSP can also be quite large, making the RSVP Path
message harder to handle.
One of the approaches defined by draft-choi-mpls-grouplabel-requirement is similar to this

approach.

6.1.2 P2P LSP Merging Approach
draft-raggarwa-mpls-p2mp-te also defines a method of grouping together P2P LSPs, but it

does this such that each P2P LSP extends right from the ingress of the P2MP LSP, rather than
from a branch node. Upstream of a branch node, there will be multiple P2P LSPs sharing a
common path.
However, the set of P2P LSPs is grouped using a modified RSVP Session Object, which has a
P2MP LSP ID field. The P2MP LSP ID allows nodes on a common path to assign the same
upstream label to each P2P LSP that is part of the same P2MP LSP.
While there are multiple overlapping P2P LSPs in the control plane, there is a single P2MP LSP
in the data plane.
• The key advantage of this approach is that because each P2P LSPs extends the full length
of the P2MP LSP in the data plane, existing fault protection/recovery mechanisms such as
Fast Reroute, error reporting, etc. all just work.
• The disadvantage is that nodes on a common path have to hold duplicate control plane
state, wasting resources. It is suggested that this can be reduced by using hierarchical
LSPs to tunnel otherwise-duplicated state directly to branch nodes, but this makes tree
modifications potentially more complex.
6.1.3 RSVP Broadcast Approach
The second of the two approaches defined by draft-choi-mpls-grouplabel-requirement uses

an RSVP broadcast-type mechanism. To date, this approach has generated less interest in the
MPLS community.
The ingress node sends a Path message to all other RSVP nodes in the network, and any node
that wants to receive multicast traffic sends a Resv message back. This draft defines RSVP Join
and Prune messages to allow egress nodes to join or leave a P2MP LSP after its initial creation.
This approach allows the ingress node to determine the set of receivers without having to rely
on an external mechanism. It is very different to the way that RSVP currently works.
6.2 PIM P2MP LSPs
draft-farinacci-mpls-multicast defines extensions to PIM to allow MPLS labels to be

distributed in a downstream unsolicited fashion along with the multicast routes.
Key attributes of these extensions are as follows.
• The P2MP LSP is restricted to following the same tree as the IP multicast tree.
• LSPs are set up even if there are no senders.
• The ingress node does not know the set of receivers, only the next-hop(s).

6.3 Tunneling IP Multicast Traffic Through an MPLS Network
There are times when it is undesirable to run a multicast routing protocol on every node within a
network, for example, it may place too much demand on the stored state.
draft-raggarwa-PIM-SM-mpls-te describes a mechanism to tunnel PIM-routed IP multicast
traffic through an MPLS network.
Key attributes of this mechanism are as follows.
• Within the MPLS network, PIM and BGP run on the Provider Edge routers only.
• The PIM routers form adjacencies to each other using the Directed Hello extensions to PIM
specified by draft-raggarwa-PIM-SM-remote-nbr. These extensions allow PIM routers to
unicast PIM messages to specific non-adjacent (in IP terms) PIM neighbors.
• Downstream Provider Edge (PE) routers send PIM Join/Prune messages to the upstream
PE router (the BGP next-hop towards the source/RP).
• The upstream PE router is responsible for creation/modification/destruction of the P2MP

LSP to the downstream PE routers.
• The upstream PE router sends a newly-defined PIM Join Acknowledge message to the
downstream PEs, containing the ID of the P2MP LSP that it will send the multicast traffic
down.
The upstream PE router could create a P2MP LSP for each source/group pair, but to improve
scalability, it is permitted to multiplex multicast traffic down the same P2MP LSP, even if this
results in downstream PE routers receiving multicast traffic when they have no interested
receivers.
For simplicity, the following example involves SSM in PIM-SM networks, but the same
PIM/MPLS interactions occur when using ASM (whether the networks represent separate
domains or subdivisions of the same PIM domain).

The following diagram shows routers R1 and R3 joining the multicast network with a source at
B. In this and in future diagrams, PE indicates a PE router.
1. R1 sends a PIM Join message to PE1.
2. PE1 creates a remote PIM adjacency to PE2 (unless such an adjacency already exists),
and unicasts a PIM Join message to it.
3. PE2 starts to set up a P2MP LSP, with PE1 as the only egress node. Note that this can
follow a different path to the PIM messages between PE2 and PE1.
Key
IP link PIM IP
PIM adjacency R1 Network A
PIM and MPLS (and BGP) 1 PE1
MPLS only
PIM only
3
2
3
S
3 R3
PE2 PE3
PIM IP PIM IP
Network B MPLS IP Network Network C

4. MPLS completes setting up the P2MP LSP.
5. PE2 unicasts a PIM Join Acknowledge message to PE1, including the P2MP LSP ID.
Key
IP link PIM IP
PIM and MPLS (and BGP)
PE1
MPLS only
PIM only
4
5
4
4
S R3
PE2 PE3
PIM IP PIM IP
6. S begins to send. The traffic flows from S to PE2 as normal, then down the MPLS tunnel
to PE1, then as normal to R1.
Key
IP link PIM IP
PIM and MPLS (and BGP) 6
PE1
MPLS only
PIM only
6
6
6 6
S R3
PE2 PE3
PIM IP PIM IP

7. R3 sends a PIM Join message to PE3.
8. PE3 creates a remote PIM adjacency to PE2 (unless such an adjacency already exists),
and unicasts a PIM Join message to it.
9. (a) PE2 starts to modify the P2MP LSP to include PE3 as another egress node.
(b) Multicast traffic flowing via the LSP to PE1 is unaffected.
Key
IP link PIM IP
PIM and MPLS (and BGP) 9b
PE1
MPLS only
PIM only
9b
9a
9a & b
9a
9b 7
S 9a & b R3
PE2 8 PE3
PIM IP PIM IP

10. MPLS completes modifying the P2MP LSP.
11. PE2 unicasts a PIM Join Acknowledge message to PE3, including the P2MP LSP ID.
Key
IP link PIM IP
PIM and MPLS (and BGP)
PE1
MPLS only
PIM only
10
10
10 10
S R3
PE2 11 PE3
PIM IP PIM IP
12. Multicast traffic now flows to both R1 and R3, and is duplicated at the branch node within
the MPLS network.
Key
IP link PIM IP
PIM and MPLS (and BGP) 12
PE1
MPLS only
PIM only
12
12
12
12 12
12 12
S R3
PE2 PE3
PIM IP PIM IP

6.4 Tunneling VPN IP Multicast Traffic Through a Provider Network
draft-rosen-vpn-mcast extends RFC 2547 Virtual Private Network (VPN) functionality to

support multicast data transport.
The key features of this approach are as follows.
• A multicast domain is a set of VPN Routing and Forwarding Tables (VRFs) associated with
interfaces that can send multicast traffic to each other. A VRF can belong to more than
one multicast domain.
• Each multicast domain is associated with an otherwise unused multicast group address
from the address space of the service provider network.
• A service provider network instance of PIM-SM runs on each PE router. These PE routers
join the service provider multicast groups corresponding to the multicast domains to which
they are connected.
For scalability reasons, this simple mechanism does not consider whether there are
actually any receivers in a particular PE router’s part of a multicast domain, and which
customer-space multicast groups that they are members of – if a PE router is connected to
a multicast domain, it will receive all multicast traffic for that multicast domain.
• Within the service provider network, there is now a multicast tree for each multicast domain
(it is recommended that this be a bi-directional tree, for scalability). This is referred to as
the multicast tunnel for the domain.
• Customer multicast data can now be transmitted across the service provider network, but it
must be encapsulated within a packet whose outer destination address is the per-domain
multicast group address (since the inner packet will have a customer site specific multicast
group address). This encapsulation could use MPLS tunnels or Generic Routing
Encapsulation (GRE).
• Each PE router runs a per-VRF instance of PIM-SM that communicates with the CE
device. This PIM-SM instance treats the multicast tunnel as a multidrop interface that
reaches all of the other PE routers that are associated with this domain.
If a single VRF is in several multicast domains, then PIM-SM requires additional

information to make the choice of which multicast tunnel to use for which customer-space
multicast groups. This information is provided by an out-of-band mechanism.
• Note that from the point of view of the CE devices, there is no change to PIM-SM or packet
forwarding behavior.

The following diagram illustrates non-overlapping trees connecting multicast VPN sites.
Key
Link to CE device
Site in colour-coded
multicast domain
Tunnel for multicast

traffic for colour-coded
multicast domain
234.0.0.1
234.0.0.2

7. MULTICAST DATA PLANE OPERATION
This chapter describes how the data plane forwards multicast data. It also discusses the
information that the data plane must make available to the control plane. The sum total of these
interactions between the data plane and the control plane is vital to the overall architecture of a
multicast router.
7.1 Host Functionality
7.1.1 Sending Data
From the point of view of an application, sending multicast data is easy. The application just
specifies a multicast IP address as the destination IP address for the packet. For Ethernet, the
sockets stack then sets the destination MAC address to a multicast MAC address that is derived
from the IP multicast address.
7.1.2 Receiving Data
A multicast datagram arrives on an interface when a multicast router sends it to that interface,
with (for Ethernet) a multicast MAC address derived from the IP multicast address. A host that
is interested in that multicast group will thus receive the packet.
In order to know whether it is interested in a particular multicast group, the host’s sockets stack
will have collated the requirements of all of the applications on the host. When a multicast
packet arrives, the sockets stack will distribute copies of the packet to all, some, or none of the
applications, as follows.
• Prior to IGMPv3, applications could only register interest in the whole multicast group.
Datagrams would typically be delivered to all applications that had opened a multicast
socket for that group on that interface.
• With IGMPv3 source filtering, a packet from a particular source may only be delivered to
those applications that have requested that the source be included (or not excluded).

7.2 Router Functionality
7.2.1 Forwarding Data
On receipt of a multicast datagram, a multicast router must first determine whether it should
forward the packet at all. In general, a router only expects to receive multicast datagrams over
the interface that leads upstream towards the source of the data or the root of the multicast tree.
A datagram that arrives over a different interface is normally dropped as it is typically
• a packet on a multidrop interface where another router has been designated responsible
for forwarding the packet onwards
• a late-arriving packet left over from before the routing topology changed
• a packet arriving via an old route during the process of changing the distribution tree of a
particular multicast group (for example, when PIM-SM shifts from a shared tree to a
source-based tree for a particular group).
Note that this is not the case for bi-directional shared trees; packets arriving on any branch of
such a tree are forwarded out over all other branches.
If a datagram is accepted for forwarding, the data plane forwards it out over a list of interfaces,
making copies of the packet as necessary.
7.2.2 Unicast Encapsulation and Decapsulation
In some shared tree routing protocols, multicast data is initially sent over unicast between the
router adjacent to the source node and the router that forms the root of the shared multicast
forwarding tree (as described in Section 4.1.4, Data Plane Interactions). In terms of data flow
this means that
• the router adjacent to the source node receives the multicast datagram, encapsulates it
and sends it to the root using normal IP unicast
• the router at the root of the shared tree decapsulates the incoming unicast packet and
forwards it as a multicast packet.
Therefore, the data plane must either be able to perform this encapsulation and decapsulation
itself, or be able to send packets for encapsulation to, and receive decapsulated packets from,
the control plane.

7.2.3 Packet Arrival Information
As described in Section 4.1.4, Data Plane Interactions, there are several situations in which
the arrival of a multicast packet may trigger communication between the data plane and the
control plane.
• In some protocols, such as PIM-DM or MOSPF, it is difficult or impossible to set up the

correct forwarding state for every source and group in advance. Therefore, the data plane
must inform the control plane when a packet from a new source or for a new group arrives
so that the control plane can set up the state correctly.
• In protocols such as PIM-SM, when a directly connected source starts sending data to a
multicast group, the data plane must inform the control plane so that it can start
encapsulating the data and sending it to the root of the shared tree.
• The control plane typically has soft state associated to a particular multicast group and
source for that group. This state is kept alive for as long as multicast data is being
received from that source for that group. Consequently, the data plane must inform the
control plane of timing information about the arrival of multicast packets.
• In certain network configurations, it is possible that multiple upstream routers may forward
multicast traffic from the same source onto the same LAN segment. Multicast protocols
such as PIM-SM and PIM-DM do not attempt to prevent this occurring. Rather, they rely
on the data plane to detect the arrival of multicast packets from different upstream routers.
This triggers the control plane to elect a single forwarder for these packets.
• In protocols such as PIM-SM, it is possible to initiate a switch from a shared tree to a

source-specific tree. To enable the control plane to decide when to do this, the data plane
needs to inform the control plane of the bandwidth consumed by data being forwarded from
individual sources over the shared tree.
• In protocols such as PIM-SM, when the control plane has initiated a switch to a source-
specific tree, it needs to know when the source-specific tree has been established so that
the source can be pruned from the shared tree. In general, multicast routing protocols do
not have any control plane mechanism to signal that a tree towards a source has been
established. Therefore, the data plane must inform the control plane when data first arrives
over the source-specific tree.

8. SUMMARY
IP multicast is an important and growing area, involving a wide and sometimes competing range
of network protocols. There are more and more areas in which people need to distribute
information to a subset of subscribers, for example, e-learning, webinars, news feeds, stock
market feeds, and radio and television broadcasts.
Hosts use the IGMP and MLD protocols to communicate their multicast group membership
requirements to adjacent routers. Different versions of these group membership protocols
provide different functionality; significantly, IGMPv3 and MLDv2 support Source-Specific
Multicast. MLD is required if using IPv6.
A range of protocols is used between routers within a domain to build spanning trees of
multicast group members, enabling efficient delivery of multicast traffic. These multicast routing
protocols have different properties, making them suitable for use in differing environments.
PIM-SM is at present the most widely used multicast routing protocol. BIDIR-PIM is not widely
used at present, but is becoming increasingly important. PIM-DM may also be used in small
networks. DVMRP was popular in the past, but is now mostly required only to interoperate with
existing DVMRP deployments.
MSDP is the protocol most widely used today for interdomain multicast routing, between
PIM-SM domains. In the future, a true interdomain multicast routing protocol such as BGMP
may provide a more scalable solution.
There are various proposals in circulation for extending MPLS to perform multicast packet
forwarding, using point-to-multipoint LSPs. None of these have yet been standardized, but as
the use of both MPLS and multicast grows, this is likely to become an important area for future
network designers.
Although multicast is not currently widely deployed within the Internet, it has been slowly gaining
influence over the last few years. The emerging next generation technologies described in this
document, together with an ever-widening area of applications and business cases, guarantee
that multicast will soon be a highly sought-after area of technology. Hence, multicast is a
strategic area for investment for many vendors and service providers.

9. ABOUT DATA CONNECTION
Data Connection Limited is the leading independent developer and supplier of (G)MPLS,
OSPF(-TE), ISIS(-TE), BGP, RIP, VPN, PIM, IGMP, MLD, ATM, MGCP, Megaco, SCTP, SIP,
VoIP Conferencing, Messaging, Directory and SNA portable products. Customers include
Alcatel, Cabletron, Cisco, Fujitsu, Hewlett-Packard, Hitachi, IBM Corp., Microsoft, Nortel, SGI
and Sun.
Data Connection is headquartered in London UK, with US offices in Reston, VA and Alameda,
CA. It was founded in 1981 and is privately held. During each of the past 21 years its profits
have exceeded 20% of revenue. Last year, sales exceeded $36 million, of which 95% were
outside the UK, mostly in the US. Even through the current severe downturn, Data
Connection’s financial position remains secure, as does its employee base: our 200 software
engineers have an average length of service of 8 years, with turnover of <3% annually.
Our routing protocols are designed from the ground up to address next generation networking
issues such as massive Internet scalability, optical routing at multiple layers, virtual routing,
MPLS and TE/CSPF, and VPNs.
Data Connection’s Multicast IP Routing products, DC-PIM and DC-IGMP, are designed to
support the functionality required by existing multicast devices, as well as to be extensible as
new features and protocols are developed.
Our products share a common architecture to facilitate integration of the family members. The
architecture allows components to be combined in a variety of ways to create routers with
specific functionality, and allows seamless integration with Data Connection’s suite of unicast IP
Routing products. These IP Routing products include implementations of ISIS, OSPF, RIP and
BGP, which can be used to provide the multicast routing protocols (e.g. PIM) with the routing
data to build multicast distribution trees.
All of the Data Connection protocol implementations are built with scalability, distribution across
multiple processors and fault tolerance architected in from the beginning. We have developed
extremely consistent development processes that result in on-time delivery of highly robust and
efficient software. This is backed up by an exceptionally responsive and expert support service,
staffed by engineers with direct experience in developing the protocol solutions.
Jon Hardwick is the senior architect and project manager for Data Connection’s multicast
routing implementations. He plays a leading role in product architecture and standards-based
development in Data Connection’s Networking Protocols Group. He has six years’ experience
in the field of communications protocols, having worked on IP Multicast, BGP, MPLS, ATM, SIP,
Megaco, MGCP, SNA and APPN.
Data Connection is a trademark of Data Connection Limited and Data Connection Corporation.
All other trademarks and registered trademarks are the property of their respective owners.

10. GLOSSARY
This section provides a brief review of some of the terminology used in the document.
Anycast RP Anycast RP is a mechanism of having more than one RP for a

PIM-SM multicast group by configuring multiple RP routers to have
the same IP.
ASM Any Source Multicast (ASM) is a model of multicast data

transmission where receivers request all data sent to a multicast
group, regardless of its source address.
Auto-RP Auto-RP is a mechanism that was used by Cisco routers in PIM-SM

v1 to determine the mapping from multicast groups to Rendezvous
Points. A central location would multicast out group-to-RP
mappings; the chicken-and-egg problem was avoided by using
PIM-DM multicast to do this.
BGMP Border Gateway Multicast Protocol (BGMP) is a multicast routing

protocol that builds shared multicast distribution trees, but where
these trees are made up of entire domains rather than individual
routers.
BIDIR-PIM Bi-directional PIM (BIDIR-PIM) is a shared tree multicast routing

protocol, based on PIM-SM but using bi-directional trees.
BSR A Bootstrap Router (BSR) is the Rendezvous Point for a PIM-SM or

BIDIR-PIM multicast tree, as (automatically) chosen by the BSR
election procedure.
CBT The Core Based Tree (CBT) routing protocol is a shared tree
multicast routing protocol.
DF The Designated Forwarder (DF) in BIDIR-PIM is the router on a

LAN that acts as the upstream neighbor for a particular RP. One
DF is elected per LAN, per RP.
DR The Designated Router (DR) in PIM-SM is the router responsible

for Register-encapsulating data from directly attached sources, and
for forwarding on behalf of directly connected receivers.
The Designated Router (DR) in CBT is like the DF in BIDIR-PIM,

except that there is a single one per LAN.
DVMRP Distance Vector Multicast Routing Protocol (DVMRP) is a source-

based tree multicast routing protocol which uses distance vector
mechanisms to determine the next upstream router on the way to
the source. DVMRP is used as the multicast routing protocol in the
MBONE.

IGMP The Internet Group Membership Protocol (IGMP) is the protocol
used between hosts and routers on an interface to indicate
membership of IPv4 multicast groups.
MBGP Multiprotocol Extensions for BGP (MBGP) is an extension to the

BGP unicast routing protocol that allows different types of
addresses (known as address families) to be distributed in
parallel—for example, IPv4 addresses, IPv6 addresses or
RFC 2547 VPN-IPv4 addresses. This allows information about the
topology of multicast-capable routers to be exchanged separately
from the topology of normal IPv4 unicast routers.
MBONE The Multicast Backbone (MBONE) is an experimental network for

multicast data that is layered on top of the Internet, allowing
multicast data to be sent by unicast.
MLD The Multicast Listener Discovery (MLD) protocol is the protocol

used between hosts and routers on an interface to indicate
membership of IPv6 multicast groups.
MOSPF Multicast Extensions to OSPF (MOSPF) is a multicast variant of the

Open Shortest Path First (OSPF) routing protocol. This protocol
uses source-based multicast trees, and distributes link state
information to allow routers to determine the next upstream router
on the way to the source.
MRIB The Multicast Routing Information Base (MRIB) is a unicast

forwarding table, used by protocols such as PIM-SM to determine
the upstream router for a particular multicast group or multicast
source/group combination.
MSDP The Multicast Source Discovery Protocol (MSDP) describes a

mechanism to connect multiple PIM-SM domains together (each
with its own domain-specific Rendezvous Point).
PIM Protocol Independent Multicast (PIM) is a set of protocols used

between multicast routers to distribute information about multicast
group membership. The protocol is “protocol independent”
because it relies on underlying unicast forwarding table information
to find the root of the multicast distribution tree, regardless of which
unicast routing protocol has been used to generated the
information.
PIM-DM PIM Dense Mode (PIM-DM) is a source-based tree multicast

routing protocol appropriate for use when the set of members of
any particular multicast group are densely distributed through the
network.

PIM-SM PIM Sparse Mode (PIM-SM) is a shared tree multicast routing
protocol appropriate for use when the set of members of any
particular multicast group are sparsely distributed across the
network.
PIM Sparse-Dense Mode Mode of operation for a router capable of both PIM-SM and
PIM-DM operation, where PIM-SM is used for some multicast
groups and PIM-DM is used for others.
PIM-SSM PIM Source-Specific Multicast (PIM-SSM) is a subset of the

PIM-SM protocol that is intended for use in pure source-specific
multicast scenarios.
RP A Rendezvous Point (RP) is the router that forms the root of a

shared multicast distribution tree in PIM-SM.
SSM Source-Specific Multicast (SSM) is a model of multicast data

transmission where the channel is identified by the <source
address, multicast group address> pair rather than just the
multicast group address.
TIB The Tree Information Base is the collection of state at a multicast

router that describes the multicast distribution trees for all of the
multicast groups.

11. REFERENCES
The following documents provide more information on the topics covered by this white paper.
All RFCs and current Internet-Drafts may be downloaded from the IETF web site at
http://www.ietf.org/.
Note that all Internet-Drafts are work in progress and may be subject to change or may be
withdrawn without notice.
11.1 Multicast Addressing
RFC 2373 IP Version 6 Addressing Architecture

RFC 3180 GLOP Addressing in 233/8
RFC 3306 Unicast-Prefix-based IPv6 Multicast Addresses
RFC 2974 Session Announcement Protocol (SAP)
RFC 2327 Session Description Protocol (SDP)
RFC 2909 The Multicast Address-Set Claim (MASC) Protocol
RFC 2730 Multicast Address Dynamic Client Allocation Protocol
(MADCAP)
RFC 2365 Administratively Scoped IP Multicast
RFC 3569 An Overview of Source-Specific Multicast (SSM)
draft-ietf-ssm-arch Source-Specific Multicast for IP
11.2 Multicast Group Membership Discovery Protocols
RFC 1054 Host Extensions For IP Multicasting (IGMPv1)

RFC 2236 Internet Group Management Protocol, Version 2
RFC 3376 Internet Group Management Protocol, Version 3
RFC 2710 Multicast Listener Discovery (MLD) for IPv6
draft-vida-mld-v2 Multicast Listener Discovery version 2 (MLDv2) for
IPv6
draft-ietf-magma-igmp-proxy IGMP/MLD-based Multicast Forwarding
(“IGMP/MLD Proxying”)
draft-ietf-magma-snoop Considerations for IGMP and MLD Snooping
Switches
RFC 2022 Support for Multicast over UNI 3.0/3.1 based ATM
Networks

11.3 Multicast Routing Protocols
draft-ietf-PIM-SM-v2-new Protocol Independent Multicast – Sparse Mode

(PIM-SM) Protocol Specification (Revised)
draft-ietf-PIM-DM-new-v2 Protocol Independent Multicast – Dense Mode
(PIM-DM) Protocol Specification (Revised)
draft-ietf-pim-bidir Bi-directional Protocol Independent Multicast
draft-ietf-PIM-SM-bsr Bootstrap Router (BSR) Mechanism for PIM Sparse
Mode
RFC 3446 Anycast-RP mechanism using PIM and MSDP
draft-ietf-mboned-embeddedrp Embedding the Rendezvous Point (RP) Address in
an IPv6 Multicast Address
draft-ietf-pim-anycast-rp Anycast-RP using PIM
RFC 1075 Distance Vector Multicast Routing Protocol (DVMRP)
draft-ietf-idmr-dvmrp-v3 Distance Vector Multicast Routing Protocol (DVMRP)
RFC 1584 Multicast Extensions to OSPF (MOSPF)
RFC 2189 Core Based Trees (CBT version 2) Multicast Routing
11.4 Interdomain Multicast Routing
RFC 3618 Multicast Source Discovery Protocol (MSDP)

draft-ietf-mboned-msdp-deploy MSDF Deployment Scenarios
RFC 2715 Interoperability Rules for Multicast Routing Protocols
draft-ietf-bgmp-spec Border Gateway Multicast Protocol (BGMP)
11.5 Multicast Signaling
draft-ietf-mpls-p2mp-requirement Requirements for Point to Multipoint Extension to

RSVP-TE
draft-yasukawa-mpls-rsvp-multicast Extended RSVP-TE for Multicast LSP Tunnels
RFC 3353 Overview of IP Multicast in an MPLS Environment
draft-choi-mpls-grouplabel-requirement Requirements for multicast service using a group
label over MPLS
draft-farinacci-mpls-multicast Using PIM to Distribute MPLS Labels for Multicast
Routes
draft-raggarwa-mpls-p2mp-te Establishing Point to Multipoint MPLS TE LSPs

draft-raggarwa-PIM-SM-mpls-te IP Multicast With PIM-SM Over a MPLS Traffic
Engineered Core
draft-raggarwa-PIM-SM-remote-nbr PIM-SM Extensions for Supporting Remote
Neighbors
draft-rosen-vpn-mcast Multicast in MPLS/BGP VPNs
RFC 2547 BGP/MPLS VPNs


Ip Multicast Explained: June 2004 Jon Hardwick Data Connection LTD

Uploaded by

Copyright:

Available Formats

Ip Multicast Explained: June 2004 Jon Hardwick Data Connection LTD

Uploaded by

Document Information

Original Title

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

Ip Multicast Explained: June 2004 Jon Hardwick Data Connection LTD

Uploaded by

Copyright:

Available Formats

IP MULTICAST EXPLAINED

Data Connection Limited

Copyright © 2004 Data Connection Limited. All Rights Reserved.

1. INTRODUCTION AND OVERVIEW ................................................................1

Copyright © 2004 Data Connection Limited. All Rights Reserved.

Copyright © 2004 Data Connection Limited. All Rights Reserved.

1.2 What is Multicast?

• it would require the source to hold a complete list of recipients

Instead, data to multiple destinations can be delivered using multicast.

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 1

In this and future diagrams

• S indicates a host sending multicast data

Flow of multicast data Hosts that aren't

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 2

1.3 History of Multicast

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 3

• Other addresses are delegated for allocation by network administrators.

In general, identification of the multicast group address is an application-level issue.

Multicast Group Membership Discovery Protocols

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 4

• In opt-out or broadcast-and-prune protocols, every router is initially assumed to want to

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 5

PIM Join messages

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 6

Data now forwarded to

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 7

Interdomain Multicast Routing

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 8

This section summarizes the content of the rest of this paper.

• Chapter 2, Multicast Addressing, describes various aspects of multicast IP addressing,

• Chapter 3, Multicast Group Membership Discovery Protocols, describes the protocols

• Chapter 11, References, provides details of references made in this paper.

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 9

This chapter discusses various aspects of multicast addressing, namely

• the division and allocation of the multicast address space

• multicast addresses with local, administrative and global scope

2.1 Multicast Address Allocation

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 10

• 239.0.0.0/8 is the administratively scoped region. The key properties of administratively

• Packets addressed to administratively scoped multicast addresses do not cross

2.3 ASM versus SSM

SSM has the following advantages over ASM.

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 11

• IGMPv3. As described in Chapter 3, Multicast Group Membership Discovery

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 12

3.1 Internet Group Management Protocol (IGMP)

IGMP Version Microsoft Windows Version

Each implementation of IGMP is required to interoperate with all previous implementations.

3.1.1 Basic IGMP operation

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 13

3.1.2 Sending Group Membership Queries

3.1.3 Responding to Group Membership Queries

3.1.4 Improving Group Membership Latency

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 14

This feature is required to support SSM.

3.2 Multicast Listener Discovery (MLD)

3.3 IGMP/MLD Proxying

Copyright © 2004 Data Connection Limited. All Rights Reserved. Page 15

3.5 Multicast over ATM