IPv6 policy and Supernational-LIRs

Gert Doering wrote:
> Hi,
> 
> On Wed, May 29, 2002 at 09:29:33AM +0000, James Aldridge wrote:
> > For IPv6, on the other hand, a supernational registry can only get a single
> > allocation, irrespective of its size and contributions to the NCC.  I don't
> > recall this policy change being discussed in the RIPE policy making forum (the
> > LIR WG) being being put in place by the NCC for the then interim IPv6 policy.
> > 
> > I am aware that there are few supernational registries and that they are a
> > pain for the RIPE NCC but this policyy change seems to work against the
> > aggregation principles we need to follow if we're not going to have the
> > routing table growth rate we've seen with IPv4.
> 
> I don't understand why "not giving out multiple IPv6 blocks" is
> "against the aggregation principles". 
> 
> Could you elaborate on this?

The old IPv6 policy gave a single /35 to a supernational registry.  This meant
that each "sub-LIR" would have to put up with, say, a /40 (enough for a mere
256 /48 assignments) and once this was used it would be unlikely that the
next /40 available from the LIR's allocation would form an aggregatable block
with the earlier "sub-allocation".  With different "sub-LIRs" having different
(national) routing policies, these multiple non-aggregatable blocks would have
to be announced to peers.  This doesn't sound like following aggregation
principles to me ;-)

However, that was the old IPv6 policy...

> Being a bit more relaxed in judging whether a multinational LIR really
> needs a "/22" (to be a bit extreme) would mimic the "IPv4 approach"
> (give out more space than usual) fairly well.

... yes, I'm sure this would help.  The supernational LIR adds a few extra
levels of hierarchy here:

	- RIR
	- Supernational LIR
	- "sub-LIR"
	- sub-LIR's ISP customers (who are either not LIRs or LIRs who don't
		meet current IPv6 criteria)
	- end-user assignments

If a hierarchy such as this could be used to justify a larger-than-/32
allocation then I don't think that big supernational registries would have any
particular problems.

James