Re: PI/metro/geo [Re: The state of IPv6 multihoming development]

[David Conrad <david.conrad@nominum.com>]

Tony,

On 11/1/02 10:40 AM, "Tony Hain" <alh-ietf@tndh.net> wrote:

> If the upper 48 bits are constant in DNS, but continually changing in
> the routing system, there needs to be a way to pass the possible set of
> topologically appropriate replacement values between the CPE routers.
> Since this protocol would inherently have to be run between
> organizations that have no trust relationship, how that be deployable?
> If it were run between the PE routers, trust is managable to a point,
> but what would prevent something like the POTS practice of slamming?

Quarter-baked idea.  How about:

<n bits identifer, reversed>.route.arpa AAAA <8-n bits routing locator>

DNSSEC signed, with the globally unique identifier "owner" holding the KEY?

Multihoming could be implemented by:

<n bits identifer, reversed>.route.arpa AAAA <8-n bits routing locator>
                                        AAAA <other provider locator>
                                        AAAA <other provider locator>
                                        ...
                                        SIG  <blahblahblah>

The border router/tunnel end point does the lookup, chooses the routing
locator to prepend to the identifier.

Hmmm.  It'd be nice if the transport layer only looked at the identifier for
checksums.

Rgds,
-drc