Multicast Service Models

IP Multicast has been deployed in various forms, both within private networks and on the wider Internet. While a number of service models have been published individually, and in many cases revised over time, there is, we believe, no high-level guidance in the form of an Informational RFC documenting the models, their advantages and disadvantages, and their appropriateness to certain scenarios. This document aims to fill that gap. This initial version of the document is not complete. There are other topics that can be included. The aim of this initial version is to determine whether this work is deemed of value within the IETF mboned WG.

The general IP multicast service model is that senders send to a multicast IP address, receivers express an interest in traffic sent to a given multicast address, and that routers figure out how to deliver traffic from the senders to the receivers. The benefit of IP multicast is that it enables delivery of content such that any multicast packet sent from a source to a given multicast group address appears once and only once on any path between a sender and an interested receiver that has joined that multicast group. A reserved range of addresses (for either IPv4 or IPv6) is used for multicast group communication. Two high-level flavours of this service model have evolved over time. In Any-Source Multicast (ASM), any number of sources may transmit multicast packets, and those sources may come and go over the course of a multicast session without being known a priori. In ASM, receivers express interest in a given multicast group address. In Source-Specific Multicast (SSM) the specific source(s) that may send traffic to the group are known in advance. In SSM, receivers express interest in a given multicast address and specific source(s). Senders transmit multicast packets without knowing where receivers are, or how many there are. Receivers are able to signal to on-link routers their desire to receive multicast content sent to a given multicast group, and in the case of SSM from specific sender IP addresses. They may discover the group (and sender IP) information in a number of different ways. They may also signal their desire to no longer receive multicast traffic for a given group (and sender IP). Multicast routing protocols are used to establish the multicast forwarding paths (tree) between a sender and a set of receivers. Each router would typically maintain multicast forwarding state for a given group (and potentially sender IP), such that it knows which interfaces to forward (and where necessary replicate) multicast packets to. Multicast packet forwarding is generally not considered a reliable service. It is typically unidirectional, but a bidirectional multicast delivery mechanism also exists.

In this section we describe general multicast building blocks that are applicable to both ASM and SSM deployment.

IANA has reserved specific ranges of IPv4 and IPv6 address space for multicast addressing. Guidelines for IPv4 multicast address assignments can be found in . IPv4 has no explicit multicast address format; a specific portion of the overall IPv4 address space is reserved for multicast use (224.0.0.0/4). Guidelines for IPv6 multicast address assignments can be found in and . The IPv6 multicast address format is described in . An IPv6 multicast group address will lie within ff00::/8.

A host wishing to signal interest in receiving (or no longer receiving) multicast to a given multicast group (and potentially from a specific sender IP) may do so by sending a packet using one of the protocols described below on an appropriate interface. For IPv4, a host may use Internet Group Management Protocol Version 2 (IGMPv2) to signal interest in a given group. IGMPv3 has the added capability of specifying interest in receiving multicast packets from specific sources. For IPv6, a host may use Multicast Listener Discovery Protocol (MLD) to signal interest in a given group. MLDv2 has the added capability of specifying interest in receiving multicast packets from specific sources. Further guidance on IGMPv3 and MLDv2 is given in .

Is this appropriate in this document? There is discussion in .

PIM-DM is detailed in . It operates by flooding multicast messages to all routers within the network in which it is configured. This ensures multicast data packets reach all interested receivers behind edge routers. Prune messages are used by routers to tell upstream routers to (temporarily) stop forwarding multicast for groups for which they have no known receivers. PIM-DM remains an Experimental protocol since its publication in 2005.

The most recent revision of PIM-SM is detailed in . PIM-SM is, as the name suggests, well-suited to scenarios where the subnets with receivers are sparsely distributed throughout the network. PIM-SM supports any number of senders for a given multicast group, which do not need to be known in advance, and which may come and go through the session. PIM-SM does not use a flooding phase, making it more scalable and efficient than PIM-DM, but this means PIM-SM needs a mechanism to construct the multicast forwarding tree (and associated forwarding tables in the routers) without flooding the network. To achieve this, PIM-SM introduces the concept of a Rendezvous Point (RP) for a PIM domain. All routers in a PIM-SM domain are then configured to use specific RP(s). Such configuration may be performed by a variety of methods, including Anycast-RP . A sending host's Designated Router encapsulates multicast packets to the RP, and a receiving host's Designated Router can forward PIM JOIN messages to the RP, in so doing forming what is known as the Rendezvous Point Tree (RPT). Optimisation of the tree may then happen once the receiving host's router is aware of the sender's IP, and a source-specific JOIN message may be sent towards it, in so doing forming the Shortest Path Tree (SPT). Unnecessary RPT paths are removed after the SPT is established.

PIM-SM can in principle operate over any network in which the cooperating routers are configured with RPs. But in general, PIM-SM for a given domain will use an RP configured for that domain. There is thus a challenge in enabling PIM-SM to work between multiple domains, i.e. to allow an RP in one domain to learn the existence of a source in another domain, such that a receiver's router in one domain can know to forward a PIM JOIN towards a source's Designated Router in another domain. The solution to this problem is to use an inter-RP signalling protocol known as Multicast Source Discovery Protocol (MSDP). . Deployment scenarios for MSDP are given in . MSDP remains an Experimental protocol since its publication in 2003. MSDP was not replicated for IPv6.

BIDIR-PIM is detailed in . In contrast to PIM-SM, it can establish bi-directional multicast forwarding trees between multicast sources and receivers. Add more...

Within a single PIM domain, PIM-SM for IPv6 works largely the same as it does for IPv4. However, the size of the IPv6 address (128 bits) allows a different mechanism for multicast routers to determine the RP for a given multicast group address. Embedded-RP specifies a method to embed the unicast RP IP address in an IPv6 multicast group address, allowing routers supporting the protocol to determine the RP for the group without any prior configuration. Embedded-RP allows PIM-SM operation across any network in which there is an end-to-end path of routers supporting the protocol. By embedding the RP address in this way, multicast for a given group can operate inter-domain without the need for an explicit source discovery protocol (i.e. without MSDP for IPv6). It would be desirable that the RP would be located close to the sender(s) in the group.

PIM-SSM is detailed in . In contrast to PIM-SM, PIM-SSM benefits from assuming that source(s) are known about in advance, i.e. the source IP address is known (by some out of band mechanism), and thus the receiver's router can send a PIM JOIN directly towards the sender, without needing to use an RP. IPv4 addresses in the 232/8 (232.0.0.0 to 232.255.255.255) range are designated as source-specific multicast (SSM) destination addresses and are reserved for use by source-specific applications and protocols. For IPv6, the address prefix FF3x::/32 is reserved for source-specific multicast use.

In this section we discuss the applicability of the ASM and SSM models described above, and their associated protocols, to a range of deployment scenarios. The context is framed in a campus / enterprise environment, but the draft could broaden its scope to other environments (thoughts?).

PIM-DM remains an Experimental protocol, that appears to be rarely used in campus or enterprise environments. Open question: what are the use cases for PIM-DM today? In campus scenarios, PIM-SM is in common use. The configuration and management of an RP is not onerous. However, if interworking with external PIM domains in IPv4 multicast deployments is needed, MSDP is required to exchange information between domain RPs about sources. MSDP remains an Experimental protocol, and can be a complex and fragile protocol to administer and troubleshoot. MSDP is also specific to IPv4; it was not carried forward to IPv6. PIM-SM is a general purpose protocol that can handle all use cases. In particular, it is well-suited to cases where one or more sources may came and go during a multicast session. For cases where a single, persistent source is used, PIM-SM has unnecessary complexity. As stated above, MSDP was not taken forward to IPv6. Instead, IPv6 has Embedded-RP, which allows the RP address for a multicast group to be embedded in the group address, making RP discovery automatic, if all routers on the path between a receiver and a sender support the protocol. Embedded-RP is well-suited for lightweight ad-hoc deployments. However, it does rely on a single RP for an entire group. Embedded-RP was run successfully between European and US academic networks during the 6NET project in 2004/05. Its usage generally remains constrained to academic networks. BIDIR-PIM is designed, as the name suggests, for bidirectional use cases.

As stated in RFC4607, SSM is particularly well-suited to dissemination-style applications with one or more senders whose identities are known (by some mechanism) before the application begins. PIM-SSM is therefore very well-suited to applications such as IP TV. Some benefits of PIM-SSM are presented in RFC 4607: "Elimination of cross-delivery of traffic when two sources simultaneously use the same source-specific destination address; Avoidance of the need for inter-host coordination when choosing source-specific addresses, as a consequence of the above; Avoidance of many of the router protocols and algorithms that are needed to provide the ASM service model." A significant benefit of SSM is its reduced complexity through eliminating network-based source discovery. This means no RPs, shared trees, SPT switchover, PIM registers, MSDP or data-driven state creation. It is really just a small subset of PIM-SM, plus IGMPv3. This makes it radically simpler to manage, troubleshoot and operate. SSM is considered more secure in that it supports access control, i.e. you only get packets from the sources you explicitly ask for, as opposed to ASM where anyone can decide to send traffic to a PIM-SM group address. It is often thought that ASM is required for multicast applications where there are multiple sources. However, RFC4607 also describes how SSM can be used instead of PIM-SM for multi-party applications: "SSM can be used to build multi-source applications where all participants' identities are not known in advance, but the multi-source "rendezvous" functionality does not occur in the network layer in this case. Just like in an application that uses unicast as the underlying transport, this functionality can be implemented by the application or by an application-layer library." A disadvantage of SSM is that it requires hosts using SSM and (edge) routers with SSM receivers to support the new(er) IGMPv3 and MLDv2 protocols. The slow delivery of support in some OSes has meant that adoption of SSM has also been slower than might have been expected, or hoped.

One of the challenges in wider-scale multicast deployment is its scalability, if it is expected that multicast-enabled routers are required to hold state for large numbers of multicast sources/groups. In practice, the number of groups a given router needs to hold state for is limited by the propagation of the multicast messages for any given group, e.g. because only a specific connected set of routers are multicast-enabled, or because multicast scope borders have been configured between multicast-enabled routers for access control purposes. Further, protocol policy/filters are typically used to limit state, as well as access control. IPv4 multicast has no explicit indication of scope boundaries within its multicast address format. The prefix 239.0.0.0/8 is reserved for private use within a network, as per , and is believed to be in common usage. Other scopes within this range are defined, e.g. Organizational Local Scope, but whether this is in common use is unclear. In contrast, IPv6 has specific flag bits reserved to indicate the scope of an address, e.g. link (0x2), site (0x5), organisation (0x8) or global (0xe), as described in . Such explicit scoping makes configuration of scope boundaries a simpler, cleaner process.

Do we want to go here, and if so which protocols should we mention? FLUTE might be one example.

Interdomain peering best practices are documented in .

Open question - do we want to look at L2 models, e.g. as might be applied at an IXP?

Anything else to add here?

Aim to add 2-3 deployment examples here, if deemed useful. Perhaps one PIM-SM/MSDP/Anycast-RP, one Embedded-RP, one SSM?

Do we wish to make a very strong recommendation here for the SSM service model, and thus for PIM-SSM, even in multi-source applications? Is this document Informational or BCP? Currently assumed Informational.

Do we need general text on multicast security here, or not?

This document currently makes no request of IANA. Note to RFC Editor: this section may be removed upon publication as an RFC.

TBC if draft progresses...