[6bone] Exchange Point Addresses

[Robert Elz]

    Date:        Tue, 18 Jun 2002 07:23:48 +0900
    From:        itojun@iijlab.net
    Message-ID:  <20020617222348.4AD6B4B24@coconut.itojun.org>

  | 	another option is to use link-local address only on IX segment.

That assumes there's only one segment, which is not necessarily
going to be true.

  | 	traceroute doesn't really have problem, as many of the routers
  | 	i know of do not do strict strong-host model.

I don't think that "strong-host model" makes sense when applied to
a router, but that's beside the point probably.

This one assumes that everything in question has a a global address,
but from where to get that address was the initial question.
Note - there's no guarantee that a router in the middle of the
exchange, which does not directly connect to any of the peering
networks, will have any other global addresses (clearly, it could
do, as you have done, use address space from one of the providers,
but avoiding that is sometimes desirable).

A non-routable global address (as others have said) should work just
fine (non-routable in this context means "non routable from the world
at large", not "non-routable from the connected networks" of course".)

It needs to be global, so packets from it can be sent to everywhere,
it doesn't need to be routable, as no-one needs to be able (from arbitrary
places) to send packets back to it.

If broken RPF implementations cause problems, they should be fixed
(refusing packets that come from a path where the route to the destination
is on another path can (in appropriate circumstances) be justified.
Refusing packets because there's no current route makes no sense for the
purposes RPF is used.

Being global, there's no problem with IP6.ARPA lookups, or anything else
like that.

kre