RE: Fwd: Minutes / Notes

["J. Noel Chiappa" <jnc@ginger.lcs.mit.edu>]

    > From: "Tony Li" <Tony.Li@procket.com>

    >> This is only important if you want TCP connections to be able to
    >> survive having an incoming link fail .. This may not be an important
    >> goal

    > I believe that the WG has come to rough consensus that this is, in
    > fact, an important goal for us to solve.

*It is architecturally crucial that the WG resolve this point.*

If you don't need this capability, then the simplest# architectural change you
really need to do widespread multi-homing is multiple addresses. (As I have
pointed out before, separation of location and identity is not really crucial
to do most multi-homing things.)

However, if you want connections to a multi-homed entity to be able to
withstand loss of connectivity to one of the addresses, then you more or less
have to have separation of location and identity.

Saying "oh we don't have to separate location and identity - we'll just use a
set of addresses, the way SCTP does, or a distinguished address, the way
MobileIPv6 does" misses the point: architecturally, even in those cases you
*have* separated location and identity - it's just that the "identity" and
"location" names are drawn from the same namespace.

All the architectural/engineering challenges - such as protecting the binding
between the two against intruders, finding the mappings back and forth,
handling error recovery, etc, etc, etc are basically the same, whether or not
the two namespaces (used for identity and location) are the same or not.

Yes, there may be engineering/architectural advantages to having two separate
namespaces (with or without identical syntax, an intermediate point) - but
that is a *separate* question.


There is also the issue of whether or not you want to carry both the location
and identity in all packets. You more or less have to have the location,
otherwise the packet can't get there. Whether or not you carry the identity
as well is an architectural/engineering question.

There are also engineering questions having to do with TCP (because of the
TCP checksum), and whether or not interoperation with unmodified TCP's is a
goal. The latter can be done, but you're probably going to have to do it with
something that uses MobileIPv6 mechanisms.

I'll stop there for now...


	Noel


#: I said "simplest because there are some other wild ways of doing
widespread multi-homing that involve substantial changes to the routing
architecture.