[ppml] [address-policy-wg] Those pesky ULAs again

Paul_Vixie at isc.org Paul_Vixie at isc.org
Fri Jun 8 13:51:15 EDT 2007


> > but since we could never possibly fit all of ipv6 into the DFZ, and since
> > the cost and availability of pi is theoretically manageable by us (the RIR
> > system) to make sure everybody who needs it can get and can afford it, i
> > fail to see the virtue of making some of it cheaper and worth less.
> 
> Well, I was really in favour of ULA-C but now given some time and after
> listning to all of the arguments, and maybe most important I realized one
> huge thing that ULA-C maybe can't provide, DNS, which leave it useless for
> our usage.
> 
> Without reverse DNS possibility ULA-C is useless.

i agree, but i think that the implications of your statement are very telling.

if these addresses really did correspond to "private" networks, then reverse
dns could be managed with cutouts, much as is done for RFC 1918 in-addrs now.

but since we all expect that these "private" networks will be interconnected
among private (non-transit) relationships, we know that the entity seeing one
of these "private" addresses and needing to know the PTR for it may not have
a direct relationship to the owner of the network using the address.  that's
another way of saying that these addresses are not actually private, or local;
the fact that they aren't intended to be carried in the DFZ matters not at all.

> On the other hand, why do we need PI? What we need is a policy that make
> sure those that could gain from ULA-C can get public routable IP space, and
> then they can themself decide if they want to route it or not. It's possible
> and upto them.
> 
> If we have to call it PI then so be it... I really dislike the name but I
> can live with it.

for this discussion, i think we should forget about the distinction between
PI and PA, which only make sense when there is a "provider."  perhaps the
automotive exchange network will be an LIR for assigning network numbers to
its members, or perhaps not.  the term we've probably been looking for is UA
(unique address, or universal address) in comparison to IPv6 "site local" or
IPv4 RFC1918 (which are nonunique and nonuniversal; sometimes "ambiguous").

everybody needs UA, whether they have a provider or not.  some UA's will be
in the DFZ and some not, according to the business needs of the network owner.
one of those business needs is "nobody will route PI UA for me" or "the cost
of routing PI UA is very high" and so, some UA will be PA UA.  that's not a
RIR issue per se, other than that it may drive policy toward smaller/cheaper
UA allocations which all come from some particular /10 so that draconian DFZ
router operators can reject these routes if they don't originate in peers.



More information about the ARIN-PPML mailing list