From: T Scott Saponas (ssaponas@cs.washington.edu)
Date: Wed Nov 10 2004 - 07:35:32 PST
Review by T. Scott Saponas
"Review of The PIM Architecture for Wide-Area Multicast Routing" presents
the Protocol Independent Multicast approach to multicast routing. The main
contribution of PIM is its ability to work efficiently for sparse and low
bandwidth multicast groups. Unlike many previous approaches, PIM is more of
an opt-in way of building multicast trees. In some schemes you periodically
flood the network with your multicast and routers send prunes if they don't
want to hear it (opt-out). Here, clients send explicit join messages up the
multicast tree. Another contribution of PIM is the use of source specific
trees. This is meant to prevent flooding of single links by all multicasts
coming from one direction.
While it does seem useful to fix the flaw that other multicast schemes are
inefficient for low-bandwidth and widely dispersed multicast, the paper does
not motivate this well. I'm left asking the question: Why not just use
unicast when your multicast is widely dispersed and low bandwidth in stead
of complicating routing? However, it does seem their opt-in scheme might
provide benefit to multicast even where multicast is absolutely needed (when
unicast would too much for the source to handle or would clog up the
network). Other limitations of this work are there is very little (if any)
evaluation/simulation of their approach reported and their still seem to
remain some unanswered questions about how PIM will address certain issues
such as how PIM will work when reservations/QOS is needed.
While this idea of opt-in and more receiver-driven multicast is appealing
from an efficiency standpoint, I would like to see PIM better thought out in
terms of its ability to meet the demands all multicast like QOS issues and
for it to be tested/simulated.
This archive was generated by hypermail 2.1.6 : Wed Nov 10 2004 - 07:35:39 PST