[ppml] IPv6 addressing plans

[Stephen Sprunk]

Thus spake <michael.dillon at bt.com>
> Does anyone have any reasoning why a network spanning two or
> more of the RIR regions, should or should not get separate ISP
> allocations from each region?
>
> I'm not just interested in opinions, but also the reasoning behind
> them, especially any technical pros and cons.
>
> In addition, are there any characteristics that define a good IPv6
> addressing plan for a network operator?
>
> We've just received an IPv6 /22 from RIPE based solely on
> projections in our European network infrastructure. Fairly soon
> we will have to decide whether to internally assign chunks of that
> space to our North American network or to go to ARIN for a
> separate IPv6 allocation based on North American needs only.

It is generally assumed, if not stated explicitly in policy, that addresses 
issued by an RIR are for use within that region only.  If the use outside 
the region is a negligible part of address use, nobody's going to complain, 
but you shouldn't be assigning space _to customers_ in the US from a RIPE 
block, and vice versa.

Also, if your network in the US is of non-negligible size, it's assumed that 
the topology is going to be mostly independent and will be a separate AS 
speaking BGP to your AS in Europe.  In that case, your US network would be a 
separate "organization", and would be no different in the RIRs' eyes than if 
your networks were two independent companies.  You probably have separate 
legal entities in such a case anyways, though that's not technically 
required for you to claim multi-organization status.  Multiple organizations 
should not "share" blocks between them, because that makes paperwork -- and 
routing -- a mess.

Another thought is that using the same block in multiple regions would 
defeat RIR-level aggregation; if RIPE-issued blocks are all exclusively used 
in the RIPE region, smaller (i.e. non-global) players in other regions can 
filter them from their routers and just point a static route for all of 
RIPE's address space "that way", i.e. towards Europe.  If, in contrast, you 
use your RIPE block in the US, people can't do that and we're doomed to 
every DFZ player having to carry redundant routes for other regions (or 
implement difficult-to-maintain filter lists) just to handle the few 
exceptions.

The main justification I see for using a block outside the region that 
issued it is if your use in the other region is not sufficient to get a 
block from the appropriate RIR.  This might be the case for a European ISP 
that buys a pipe to show up at a US IXP, or vice versa; addressing a couple 
routers is not enough to get a block from the "correct" RIR, but it makes 
sense to use addresses from the "wrong" RIR in such a case.  Likewise, if an 
end-user org has a PI assignment from a certain region, it makes sense that 
all their sites worldwide would use addresses from that block _if they did 
not have connectivity to the outside world except through the issuing 
region_.  For instance, if a office in Moscow has to go through HQ in the US 
to reach the Internet (which is not unusual), they should be addressed from 
that company's ARIN block; if they got a second connection in Europe, the 
sites behind that connection should be renumbered to RIPE space.

S

Stephen Sprunk         "God does not play dice."  --Albert Einstein
CCIE #3723         "God is an inveterate gambler, and He throws the
K5SSS        dice at every possible opportunity." --Stephen Hawking