All things related to multicast

[Bill Fenner <>]

>except that it's hasn't in practice because we have two things that w e 
>than have to evolve in parallel.

Indeed.  However, look at how long it's taking to get from IGMPv2 to IGMPv3
stacks.  At the time that IGMPv2 was the "sweet spot" of IGMP, i.e. the
time that all that implementation and deployment was happening, DVMRP was
the "sweet spot" of routing protocols -- PIM was this bright new opportunity
poking its head over the horizon and MOSPF was much more well developed.

If the host<>router communication was the routing protocol, all of the
end routers would *still* be trying to talk DVMRP to the hosts, and there
would be a tiny minority that talked PIM, even though PIM is ubiquitous
in the network and there are no significant bits of DVMRP routing.

Now, just because it made sense in the past doesn't mean that it still
makes sense now -- maybe PIM is the end-all and be-all and that's all we
should care about.  But it seems unlikely to me that
a) revving routing protocols in the host is any easier than revving IGMP
   in the host.
b) PIM is the only routing protocol we ever want to have.

Maybe the problem with IGMP was that it was implemented in the kernel,
and we should have let the kernel collect the membership information,
reference-count sockets and program the multicast filter and such but
then it should have notified a daemon of the memberships and the daemon
could talk whatever protocol is needed.

  Bill