Re: [udlr] Questions

[Steve Deering <deering@parc.xerox.com>]

> 1) why not send the RIP packets in multicast to multiple feeds in case you
>    have many of them serving the same subnet.

Probably more serious and more important than the lots-of-senders problem,
above, is the lots-of-receivers problem.  Since I wasn't able to attend the
BOF in Montreal, I don't know if this was discussed, but I didn't see it
in the minutes or the list of questions.  It seems quite likely that
direct broadcast TV satellites will be used to deliver IP packets to
potentially millions of homes, but the approach suggested in the INRIA
drafts of having every receiver terrestrially unicast (or multicast)
periodic routing messages back to the feed(s) will create a scaling problem
at the feed(s).  One solution is to forget about the back-channel routing
updates, and simply statically configure the feed(s) to know that a given
prefix (or small set of prefixes) identifies destinations reached over the
satellite channel.  However, that imposes undesirable addressing constraints
on the home subnets -- it's basically another example of "provider-based"
addressing.  Has anyone got any good ideas how to avoid route implosion
at the feed(s) when there are millions of receivers?

> 3) why don't run routing protocol even in the case of basic access?

Yes, this is certainly a question I have.  The single-node-at-home scenario
might be common initially, but it seems likely to become irrelevant in a
very short time -- homes will contain multiple IP devices on one or more
home subnets, all connected to the outside world by a router (perhaps
disguised as a "set-top box") or two.  Requiring that the node attached to
the satellite receiver must run the routing protocol still permits the
single-node-at-home scenario, as the degenerate case of a router that serves
only itself.

Steve