From: Tom Christiansen (tomchr@ee.washington.edu)
Date: Tue Nov 09 2004 - 18:41:03 PST
This paper presents a protocol for multicasting across the Internet.
Previous multicast/broadcast protocols transmitted data to everyone on the
network. Unless a majority of network hosts were subscribing to the data
broadcast, a significant waste of bandwidth occurred as data was
transmitted down branches of the network where there was no receivers. This
protocol (PIM - Protocol Independent Multicast) solves this problem through
a join/prune mechanism. If a branch of the network has no recipients for
the data flow, the router for that branch sends a prune message back to the
broadcaster and until timeout occurs, this router is then taken off the
broadcast list. This seems like a pretty good idea.
Another contribution made by this paper is a comparison study of
center-based trees vs. SPT (Shortest Path Tree, I assume). Unfortunately,
the graphics were degraded significantly in the pdf conversion, and the
text is a bit inconclusive (there are pros and cons of both tree types).
One concern with this protocol is the rather obvious way to create a denial
of service attack. Simply send a prune message to the broadcasting hosts
with a fake IP address and you can disconnect significant portions of the
Internet from the transmitter. - Along the same lines, if the protocol is
used to broadcast to a very low number of users, it seems like there would
be a flood of prune messages from parts of the network with no recipients
for the broadcast. This might create a denial of service condition similar
to the one caused by a TCP SYN flood attack.
The explanation of exactly how the protocol works is a bit weak. Adding a
better state diagram would probably have helped.
This archive was generated by hypermail 2.1.6 : Tue Nov 09 2004 - 18:41:13 PST