How to choose IPv6 addresses for customer links?

[Pekka Savola]

On Fri, 30 Jan 2009, Martin Horneffer wrote:
> Consider a service provider that provides IPv6 services to leased 
> line customers. In almost all cases the customer gets a /48 out of 
> the aggregate of the service provider. In many cases and probaly in 
> most future-oriented cases the physical interface is some kind of 
> ethernet (10/100/100/10000 Mbit/s). Thus the link to the customer 
> needs its own addresses.

Actually, the link already has link-local addresses.  You could just 
run it without any global address.  But that has a caveat for another 
reason. When you need to configure static routes towards the customer, 
you're kinda stuck because you'd need to configure the nexthop to be a 
link-local address and that would not be fun for many reasons, 
including changes in CPE (MAC address change).  Otherwise link-local 
would be great.

FWIW, we're using an address out of customer's block; they can choose 
but the customers are organisations so this is not a provisioning 
problem.

We also advertise those p2p addresses to our core.  This has some 
tradeoffs.  The good is that the link is pingable even if the address 
aggregate would happen to change so that the p2p link is no longer 
covered, and that if a customer is multihomed, pinging the p2p link 
from different places would cause non-deterministic routing.  The bad 
is that for multihomed customers, strict uRPF requires a workaround. 
(See S3.2 of draft-savola-bcp84-urpf-experiences-03 for more.)

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings