[ppml] v6 Multihoming (was Re: IPv4 "Up For Grabs" proposal)

[Christopher Morrow]

On 7/11/07, William Herrin <arin-contact at dirtside.com> wrote:
> On 7/11/07, Christopher Morrow <christopher.morrow at gmail.com> wrote:
> > Bill's making a giant leap of faith that all multi-homing won't be
> > done like ipv4 multihoming and that people will be 'as good' as they
> > are today wrt de-aggregation... This seems, based on past history,
> > like a very, very bad bet.
>
> Chris,
>
> Actually I'm not. I haven't personally checked the numbers, but I'm
> told that if every org that has one or more prefixes announced in IPv4
> announces exactly one prefix into IPv6, the table would have in the

I think this depends upon how you define an org actually, but given
that some folks are doing that by 'if you have an ASN you get a /32 or
/48' perhaps the number is as low as 45k (rounded up today). There are
certainly some of these cases that may need more than one prefix, or
may have disjoint AS's that would require more than 1 prefix, double
to 90k. Add some fuzz for the multi-national corps that have
multihomed offices all over creation on 'whatever convenient provider'
is in region, add easily another 30k... This is a tough problem to
model today. Multihoming is becoming more prevalent, not less
(somewhere near 30% of uunet customers today are multihomed, and
growing).

Also, add in the (as mentioned before) existing 220k v4 routes, look
at the growth curves (see vaf at cisco.com/schiller at uu.net/geoff at apnic
presentations on same) and in 5 years you're looking at over 500k v4
routes alone.  Look at the reasons for deaggragation, look at the
update-rate for the global table...

> neighborhood of 50k-60k entries. More, deaggregation is hard to get
> away with when almost everybody has exactly a /48 and nobody has less.
> Not sayin' it won't happen, just that it won't be as common

sure, look at the data :( it doesn't have to be so much 'more
deaggragation', just more routes for all the normal reasons.

>
> Will the IPv6 DFZ eventually grow past 100k entries? Sure. Soon? No.

read the vaf/schiller presentations from RAW/RAM/IETF/NANOG... looks
like it's coming sooner than you'd think. (based on some projects that
seem to hold fairly well so far)

-Chris
(healthy disclaimer, i work with schiller at uu so I have some both
interest in his work and have presented some of it)