Re: Multicast IP

Scott Brim (swb1@cornell.edu)
Sat, 11 Feb 1995 15:23:13 -0500


For multicast routing, a good place to start is Steve Deering's thesis,
which is highly readable and covers all the basics. Look on
gregorio.stanford.edu, or perhaps in
ftp://parcftp.xerox.com/net-research/. Once you're done there, look at
the IDMR (Inter-Domain Multicast Routing) internet drafts
(draft-ietf-idmr-*), and also at the Nimrod architecture draft
(draft-ietf-nimrod-*). Those will let you know where things are
currently.

Multicast routing is not a solved problem by any means when you go
beyond LAN environments. We've got enough pieces nailed down to
deploy, but I honestly don't think that what we are doing now will
scale to millions of simultaneous groups. An important idea, which
hasn't been followed up on recently, is localized dynamic changes in
algorithm in response to changing conditions. For example if a source
only sends out one message a week, it's economical for routers just to
send it everywhere and let the end networks throw the message away,
rather than maintaining a lot of state information about where it
shouldn't go (useless 99.99% of the time). If the same source starts
sending out a message every 10 minutes, then some (but not necessarily
all) parts of the Internet might want to start maintaining state
information to make sure it only goes where it should. And so on with
similar tradeoffs. Happy reading!

...Scott