From: Craig M Prince (cmprince@cs.washington.edu)
Date: Wed Nov 10 2004 - 07:36:57 PST
Reading Review 11-10-2004
-------------------------
Craig Prince
The paper titled "The PIM architecture for Wide-Area Multicast Routing"
attempts to solve a major problem with multi-cast routing which is how to
effectively send multi-cast over a network where most of the people don't
want your messages, yet you still want to provide multi-cast.
Specifically, the network is large and the members of the multi-cast group
are sparsely dispersed across the network.
Their solution focuses on the fact that in previous multicast schemes
subscriptions and routes were setup from the sources to the receivers.
This "receiver discovery" results in many network routers having to keep
un-needed state since they had no subscribers. Instead the authors opt for
a receiver-to-source discovery system. This way the routing tree is built
up from the leaves to the root instead.
The paper does a good job of analyzing the various design tradeoffs with
respect to tree shape, etc. Indeed, one of the subgoals was to make their
protocol effecient to different traffic patterns and usages. As such the
protocol supports multiple types of tree structures (SPT, shared tree).
One concern with this paper is that while they present a protocol, they
don't provide much simulation of its effectiveness. That is, they don't
really prove that it solves the problems they set out to solve. In
addition, they don't really explore the performance limitations of their
scheme. Finally, I was concerned that their scheme (as well as all the
others mentioned) require some sort of modifications to internet routers
today. This worries me since I do not see an easy method of deployment for
this multi-cast scheme.
Overall, I thought the paper was interesting and addressed a real problem
that has limited the acceptance of multi-cast systems.
This archive was generated by hypermail 2.1.6 : Wed Nov 10 2004 - 07:36:58 PST