Re: Ohta-san's draft

[Jim Bound <seamus@bit-net.com>]

8+8 is not needed at all or differential of location and identification to
achieve multihoming.  I do think some intelligence at the end system can
help alleviate carrying routes at the borders.  That can be set as a req
per Brian's taxonomy and then done in many different forms.  Using SCTP,
Homeless Mobile IP taxonomy of altering the transport endpoint at the IP
layer, or hacks to the socket layer, libraries, or the IP layer (e.g.
bump-in-the-bump models).

Note-As stack builder I would never use bump-in-the-bump models for
production nodes that are long lived at customers.  For small nodes its
also too much extra code. I would be better off doing good job with SCTP
or Homeless Mobile IP model.  

As far as destination address selection I don't agree with the idea of
searching dns entries or having to change DNS.  It should be done based on
a routing table and to speed things up a preference metric all things
being equal.

That being said I think the architectural limitation to the predominant IP
host of tomorrow will be the routing table again (e.g. Phones, PDAs,
Embbed Systems, Internet Appliances, and Remote Control Devices).  Also I
believe IPv6 Neighbor Discovery could have options to extend the
intelligence of end nodes for Multihoming e2e.  I also think we cannot
alter the use of ND at this point either but we can extend it as the
mobile IP folks are doing for variant handoff models.

None of the above breaks the existing IPv6 stack or the APIs but adds
extensions.  

/jim