[ppml] /48 vs /32 micro allocations, answer: /48

[Paul Vixie]

> So we may need to have a conversation around that definition for IPv6.

yes.  in particular, an exchange point would actually need a /48 not a
/128.  (a /64 won't do, since exchange points now mostly use multiple
VLANs to segregate things like multicast from things like unicast.)

> Of course, there are other questions within the narrow "root
> operator" definition.  Does each IPv4 anycast root get its own IPv6
> allocations?

according to www.root-servers.org, that is indeed what's happening today.

> I'm glad to see general use of "routing table slot" as an
> economic concern.  large-but-finite = scarce

the thing is, in ipv4 there was an equilibrium to be established, and it's
moved several times.  the minimum allocation size was raised when there
were too many routes, then lowered when there were too many multihomers who
couldn't qualify for the larger size.  in ipv6 there is no equilibrium,
the smallest block we can imagine allocating is large enough for almost
anybody except perhaps some multinationals.