[c-nsp] IPv6 subnets for point-to-point links

[sthaug at nethelp.no]

> > > There's a distinct chance that a 64 bit host ID will eventually become 
> > > important to do automatic IPSEC ("crypto generated addresses") and 
> > > you'll really want to avoid having to renumber all links then.
> > 
> > Maybe you could point us in the direction of this work? I would need good
> > arguments to start using /64 on point to point links.
> 
> Well, the first stop is of course RFC3587, which very clearly says
> 
> " [ARCH] also requires that all unicast addresses, except those that
>   start with binary value 000, have Interface IDs that are 64 bits long"
> 
> (ARCH is RFC3513)

I'm afraid this is not a particularly compelling argument (to *me* at
least) for why I should waste a /64 on every point to point link, as
long as /126 definitely work.

> As for the CGAs, I think the most important two drafts/proposed standards
> are:
> 
> draft-ietf-send-cga-07.txt           --> now RFC3972
> draft-haddad-mip6-cga-omipv6-04.txt

These drafts/proposed standards apply to Secure Neighbor Discovery and
to Mobile IP. I don't see the relevance to my point to point links (I
neither want nor need any kind of neighbor discovery here - everything
is explicitly configured).

I think I'll stick to /126 for now.

Steinar Haug, Nethelp consulting, sthaug at nethelp.no