[arin-discuss] IPv6 Provider Woes

[Paul Vixie]

> > if the internet community is going to abandon CIDR and go 
> > back to non-hierarchical routing, it should be done 
> > explicitly -- eyes wide open, and not incrementally -- in 
> > denial -- through the back door.
> 
> What if I had said "receives its own ASN and /48"?  Does that fully
> address your objection about non-hierarchical routing?

no, and that's a good point.  in terms of address space conservation,
1/4294967296 and 1/281474976710656 are indistinguishable fractions of the
total when allocated no matter what the purpose.  however, both take 1 slot
in the global routing swamp, and the thought implicit in the current
allocation policy was that a /48 was 64K LANs which is basically one
well-VLAN'd campus whereas a /32 is 4B LANs which is basically 64K campuses
each having 64K LANs.  the cost:benefit to the other denizens of the global
routing swamp is much better for a dense /32 than for a dense /48 or a
sparse /32.  (where "dense" probably means 100 out of 64K possible
suballocations.)  perhaps a /32 for a each datacenter can make sense in the
current allocation policy framework if each is multihomed and each is
multitenant.  but if these are single homed or single enterprise, i think
the community's consensus will be something like "please build yourself a
backbone, or use provider-assigned (hierarchical) space."

note that f-root has an IPv6 /48 out of which we use one address.  this is
tolerable to the community only because the community needs DNS root service
and because it's one routing table slot whether it's a /48 or a /128.  this
also means that folks who filter on a /32 boundary won't see f-root in IPv6.

tony li told us at ARIN XX that until there was market pressure around
routing table slots, our allocation policies were going to be skewed, and
this sure sounds to me like what i thought he meant.