next up previous contents
Next: Transport Protocols Up: Multicast in the Internet Previous: The Multicast Model

A Brief History of Trees

In Chapter three, we discuss multicast routing topologies at great length. Here we give a brief overview of the evolution of multicast to date. Originally, the multicast routing was initially based on the thesis work of Steve Deering [Deering, 89] and comprises two parts:

Tunnels - these are used to glue together Internet sites which have multicast capable routers, but are separated by routers that do not support multicast, thus forming a virtual topology on top of the underlying Internet unicast routing. This has proved invaluable in this (and other) work in terms of deploying new versions of the routing. This has given rise to the term Mbone, short for Multicast Backbone, which is the name for the virtual glue of routers that interconnect islands of native IP multicast together via tunnels through non-multicast capable sections of the Internet.

Distance Vector Multicast Reverse Path (DVMRP) routing - this is the actual routing protocol, which is a natural extension to the age-old Routing Information Protocol, using the paths that are calculated to get from a set of sources D to a particular destination, S for unicast, as a way to get multicast from S to D.

DVMRP employs a scheme called pruning to eliminate branches from the network that do not contain members of a group whenever a source starts sending. As the Internet multicast capability has grown (in early 1998 it had several thousand sites, where a site might be a research laboratory or university Campus, in 22 countries, internationally), there have been many groups that are small, and sparse. This has meant that the amount of routing control overhead from pruning traffic (and multicast traffic delivered unnecessarily to sites without members before they are pruned) has caused people to re-think the routing scheme. Several alternates have emerged:

MOSPF (Multicast extensions to the unicast routing scheme OSPF) allows aggregation of traffic and groups and also permits paths to be chosen based on different Types of Service

CBT (Core Based Trees) is based on a manager placing a Core or center router appropriately in the network by calculating where the place is that all routing traffic would go through if we formed a minimal spanning tree from the center to all groups. This is a tricky calculation, and takes smart heuristics. It results in lower link usage than DVMRP, and does not need pruning, but it can increase the delays in paths between users, which may be critical for some kinds of multimedia interactions.

PIM (Protocol Independent Multicast) is based on a mix of the ideas from CBT and DVMRP, and relies on the underlying unicast routing to calculate its paths. It is also capable therefore of exploiting underlying policies concerning routes, including potentially, Type of Service selection of paths.

The two extremes of multicast tree topology are explained at great length in chapter three, and for a quick sneak preview, take a look at the figures [*] and 3.2 there. At the time of writing, there is a great deal of research and development in the area of multicast routing, and it remains to be seen what the main scheme will be. However, the power of the basic original IP multicast model is in no doubt.

next up previous contents
Next: Transport Protocols Up: Multicast in the Internet Previous: The Multicast Model