Border Gateway Multicast Protocol/GUM

The `Border Gateway Multicast Protocol' (previously known as GUM) was presented by Dave Thaler and is defined in draft-ietf-idmr-gum-00. GUM is one of a family of protocols needed for multicast routing and address allocation:

• MASC associates group prefixes with selected domains in an aggregatable way.
• BGP4++ propogates the multicast RIB of soure and group prefixes
• GUM constructs group shared trees of domains, in a bidirectional manner. The tree for group Gis rooted at the domain advertising the route to G.

The main features of GUM are that:

• No state or data where no sources/receivers exist
• Bi-directional tree to minimize third party dependence
• No source specific state except where required (this solves problems due to low rate sources such as RTCP)
• Source specific branches allowed if required (this relieves problems due to high rate sources)
• Group state is aggregatable for consecutive groups

Discussion of this protocol followed. It was noted that it is necessary to encapsulate packets between border routers in some cases, to make RPF checks work (this depends on the way the intradomain routing protocol works, DVMRP needs it, for example). This is inefficient, so if a high data rate is present it is possible to switch to a shortest path rather than a shared tree, which will avoid having to encapsulate (ie: a source specific branch).