From: Ethan Katz-Bassett (ethan@cs.washington.edu)
Date: Tue Nov 09 2004 - 21:04:41 PST
In this paper, the authors present their Protocol Independent Multicast
(PIM) architecture for wide area multicast. The paper first gives a good
background on distance-vector based multicast and link-state based
multicast, describing these existing mechanisms and highlighting some of
their limitations. They also present the core based tree (CBT) protocol and
describe their simulations of its performance. One of their findings was
that CBT gave up to a 40% increase in max delay, compared with SPT; that
amount did not strike me as being particularly high.
The existing methods did not scale well when multicast group membership is
sparsely distributed across the network, and the authors design PIM to
address this problem. Under sparse conditions, new members join a PIM tree
by sending a PIM join. PIM also includes both shared trees and SPTs,
allowing the advantages of each under appropriate circumstances. As the
name implies, PIM does not depend on the underlying unicast routing
protocol. All these considerations seem appropriate, letting PIM perform
well under different group densities and over different protocols.
I thought that the paper could have been clearer; it was not that easy to
get a good sense of the protocol. It included no PIM performance
measurements or analysis of performance; this seemed particularly unusual
since the paper presented simulations of CBT, an alternative to PIM. The
authors might have included, for instance, measurements about how
reverse-SPT tends to perform compared to SPT under "normal" network
configurations/conditions.
This archive was generated by hypermail 2.1.6 : Tue Nov 09 2004 - 21:04:47 PST