From: Karthik Gopalratnam (karthikg@cs.washington.edu)
Date: Wed Nov 10 2004 - 10:00:45 PST
Review 19 - PIM
This paper provides an alternative multicast scheme to MOSPF and DMVRP,
called Protocol Independent Multicast, where the primary aim is to
efficiently handle groups that are spread across WANs.
This paper addresses the issues with efficiency in existing multicast
schemes, which are dependent on some form of IP routing scheme. These
schemes are good for dense groups, but for groups that are spread out, many
routers on the network end up doing a lot of useless work in maintaining
soft state, handling pruning messages and processing IGMP packets even if no
group members belong to the domain handled by that router. This paper,
instead of using the single-SPT-tree method used by these schemes, proposes
using source specific trees, which can provide far better guarantees in
terms of efficient sparse group support, high quality data distribution,
routing protocol independence and robustness.
While the ideas presented in this paper make sense, there are various
issues that I'm not comfortable with. Firstly, it appears that the design of
the timeout is not really well motivated, and seeems arbitrary. Unless the
timer is really designed well, the network would end up sending probably the
same number of prune messages as the original protocol. It would seem that
getting this issue just right is central to the success of this protocol.
Secondly, the paper presents no simulation results at all to validate this
approach. It would have been good to see some numbers on how much more
efficient this is in terms of the routers doing less work, as compared to
MOSPF and DMVRP.
This archive was generated by hypermail 2.1.6 : Wed Nov 10 2004 - 10:00:49 PST