Section 15.3. Interdomain Multicast Protocols

15.3. Interdomain Multicast Protocols

Interdomain multicast protocols are designed for hierarchical and Internet-wide multicast purposes. Within a domain, a network manager can implement any routing protocol desired. The challenge in interdomain multicast administration is choosing the best external link to route to hosts in an external domain. Among the protocols in this category are multiprotocol Border Gateway Protocol (MBGP), Multicast Source Discovery Protocol (MSDP), and Border Gateway Multicast Protocol (BGMP).

15.3.1. Multiprotocol BGP (MBGP)

Multiprotocol BGP (MBGP) is an extension to its unicast version, Border Gateway Protocol (BGP). In Figure 15.8, three domains are connected through two types of paths: a unicast path handled by BGP and a multicast path handled by MBGP. With BGP, the multicast routing hierarchy operates the same way as multiple unicast routing does. Between each two domains, two corresponding border routers compute the set of domain parameters that should be traversed to reach any network. Typically, the parameters in one domain are not known or trusted by the others.

Figure 15.8. MBGP: interdomain multicast routing

Each domain-management system advertises the set of routes that can reach a particular destination. Packets are routed on a hop-by-hop basis. Thus, MBGP can determine the next hop to a host but cannot provide multicast tree-construction functions. Sources within a domain register with a rendezvous router, and receivers send joins to this router. This join message is required to find the best reverse path to the source.

This protocol can handle multiprotocol paths, and each participating router needs to know only the topology of its own domain and the paths to reach each of the other domains. Since MBGP is an interdomain protocol, a class D address is not carried in an MBGP message, and it does not carry information about multicast groups. Because the time and frequency of join messages transmission are not determined in this protocol, and multicast trees in various domains are not connected, other multicast protocol choices are more attractive.

15.3.2. Multicast Source Discovery Protocol (MSDP)

One of the main issues in interdomain multicasting is how to inform a rendezvous router in one domain while there are sources in other domains. In other words, rendezvous routers of two adjacent domains are not able to communicate with each other when one receives a source register message. In practice, only one rendezvous point is in each domain. This issue becomes more critical when group members should be spread over multiple domains. In this case, multicast trees among domains are not connected. As a result, the traffic can reach all receivers of a domain, but any sources outside the domain stay disjoint .

The Multicast Source Discovery Protocol (MSDP) has potential solutions to these issues. Figure 15.9 shows how this protocol operates. A unique feature of this protocol is that it assigns representatives in each domain. A representative reports to other domains the existence of active sources. A new source for a group must first register with the domain's rendezvous router.

Figure 15.9. MSDP: interdomain multicast routing

Each MSDP representative in the domain detects the existence of the new source and sends a message to all MSDP representatives, as shown. The representatives check whether the broadcasting representative has sent the message through the correct path to prevent possible message looping. Once this process is complete and the message is on the correct router interface, the message is forwarded to all remaining associated representatives. An MSDP representative and the rendezvous router within a domain can be the same point; this is the checkpoint where the state of group membership is checked. If it has the correct state, the router sends a join message to the source address mentioned in the message. Here, we assume that the intradomain multicast is performed using PIM, whereby a join message is generated and processed . The message is then forwarded on the multicast tree by the rendezvous router; once all group members receive data attached to the message, they may use a shortest-path tree, using sparse-mode PIM. This process ends when all the representatives finish the process.

This multicast procedure uses a combination of three protocols: sparse-mode PIM, MBGP, and MSDP. Although this multicast procedure has been well accepted as a practical multicast method, its complexity results in a timing issue. One aspect is scalability. Also, when sources are bursty or group member join and leave events frequently, the overhead of managing the group can be noticeable. This fact in turn creates the timeout issue, whereby the period of silence between packet bursts becomes greater than the forwarding-state timeout. MSDP solves this issue by selecting and processing every n packets in burst. However, this trick is not quite a practical solution when a fairly large network is under multicasting.

15.3.3. Border Gateway Multicast Protocol (BGMP)

The Border Gateway Multicast Protocol (BGMP) is based on the construction of bidirectional shared trees among domains using a single root. Finding the best domain to place the root of such shared trees is a challenge, but several solutions are available. One of the methods of address resolution is the Multicast Address-Set Claim (MASC) protocol, which guarantees the immediate resolution of address collisions.

Another method of resolving this issue is to use root-addressed multicast architecture (RAMA) by which a source is selected as the root of the tree. This way, the complexity of root placement in other multicast routing protocols can be eliminated. RAMA is of two types. The first type is express multicast : The root of the tree is placed at the source, and group members send join messages on the reverse path back to the source. This protocol is aimed at systems that use logical channels, such as single-source multimedia applications, TV broadcast, and file distribution. The second type is simple multicast : Multiple sources per group are allowed. In this case, one source must be selected, and the root of the tree is put at this node as a primary and first-hop router. Then, all receivers send join messages to the source. The next step is to construct a bidirectional tree through which additional sources send packets to the root. Since a bidirectional tree is constructed , packets arrive at a router in the tree and are forwarded both downstream to receivers and upstream to the root.