Re: [v6ops] draft-ietf-v6ops-enterprise-incremental-ipv6 WGLC

[Lorenzo Colitti <[e-mail]>]

On Tue, Aug 6, 2013 at 11:53 AM, Victor Kuarsingh <[e-mail]> wrote:

> What about the use case of a host that has no global connectivity
> requirements? I know of cases in the SP world where this is true, although
> this draft is for Enterprise, but they may have similar cases (I.e. Closed
> network environments, testing beds, etc).
>

This case is neither as simple nor as clear-cut as it sounds.

If a host truly has no global connectivity requirements then it doesn't
matter what IP addresses it uses as long as they are globally unique. (They
still need to be globally unique because even though the host doesn't talk
to the outside world, it will talk to something that [talks to something
that...] talks to the outside world).

In that case, then that host can use any address that's globally unique,
including global unicast or ULA. Global addresses have the advantage that
they are unambiguously globally unique, whereas ULA addresses are only
probabilistically unique. This means that if you have enough ULA to cause a
collision then you will have problems, and if someone misconfigures their
machines with the ULA prefix that you picked, then you're sort of SOL. On
the other hand, ULA has an advantage over PI addresses because the latter
may need to be renumbered if you change ISPs.

There's also the point that when people say "will never talk to the outside
world", that's more often than not an oversimplification. :-)

Since these points are pretty subtle, they should be carefully worded. I
think the right place for that is in the ULA usage document. Why not simply
punt to that document?