[ppml] APNIC policy proposal to create a regulated market in IPv4 addresses

Paul Vixie wrote:
>>> Also, does anyone believe ISPs are going to remove the filters that  
>>> implicitly limit the length of a usable (for the purposes of highest  
>>> likelihood of global routability) IPv4 prefix to a /24?  If so, why?
>> I hope not, and if subdivision/deaggregation gets bad enough we will see 
>> filters tighten.  In a worst case scenario we filter at /20 across
>> the board  and anyone using longer prefix gets to move.
> 
> what would you have f.root-servers.org do in that case?  our IPv4 address
> is in a /24, and our IPv6 address is in a /48.  even apart from the question
> of how long it takes to renumber a root name server, where would you like us
> to move?  we have other address blocks but those other blocks aren't anycast.

I put that out there as a worst case scenario, as in my sessions won't
stay up without it.  Surely other less drastic measures would be used
before that point, but new routes would always pop up to work around
each incremental step.  This is all in the context of an essentially
unregulated market for IP addresses where arbitrary subdivision is
possible.

Sounds pretty grim right?  It is.  I would make exceptions for
*.root-servers, but don't expect everyone to.  People are lazy,
Many will copy templates they find from various sources, each with
different exceptions and different typos.  For a contemporary example,
http://www.space.net/~gert/RIPE/ipv6-filters.html STILL lists 2620::/23
with a recommended filter size of /32.

The good news is, even in this catastrophic scenario, d and h-root are
fortunately within /16 assignments and can give out new addresses for
the rest for those with outdated hint files.

The bad news is of course for everyone who would have to renumber,
possibly multiple times as filters are incrementally tightened.

Those who suggested filtering on minimum RIR assignment sizes are
implying that it is not possible to subdivide assignments in the
forthcoming market.   This is exactly the point I was trying to
make, that allowing subdivision even below /20 will eventually
result in drastic filtering.

- Kevin