Re: A new spin on multihoming: multihoming classes.

[Iljitsch van Beijnum <iljitsch@muada.com>]

On Thu, 6 Sep 2001, Randy Bush wrote:

> > Many of the large networks would really like an 8k global routing table
> > in IPv6. That means: no more than 8192 ISPs with a PA block of their own.

> really?  as it forms the base of your argument, i must ask if there is a
> basis for this assertion?

I've read words to that effect on this list, and I may have read the same
thing in a document (draft, RFC, RIR document), but I'm not absolutely
sure on the latter.

> my impression is that large (and small) providers would like fairly stable
> routing which converges predictably in reasonable time.  if that can be done
> with ten, 100k, or 5,000,000 routes is not so important.  but, as i said,
> that is merely my impression.

What are you saying? That there is no problem? That BGP implementations
need to get better and stop muliplying flaps? Or something else?

If we want an 8k routing table, we can't give ISPs more than a hand full
of PA blocks and we have to make sure only real ISPs get them. Not
limiting the number of PA blocks per ISP probably won't hurt much: with
the larger blocks they get with IPv6 they won't burn through them as fast
as in IPv4. If we relax the "real" ISP rule, there will be some more
growth, since many enterprises will become "ISPs" to be able to multihome
without problems. But this is pretty expensive, so the growth of the
global routing table should still be fairly reasonable. if we allow PI
and/or /48s out of PA space in the DFZ without any restrictions, the
routing table will grow as we've never seen before, and it is very likely
that many of the new networks will have less stable connections than
current networks and the new, large routing table will also be less
stable.