Re: draft-ietf-multi6-multihoming-requirements-06.txt

[Jay Ford <jay-ford@uiowa.edu>]

On Wed, 18 Jun 2003, Craig A. Huegen wrote:
> > Call me a control freak, but I (as campus network routing guy) would like
> > to get a little more determinism than is afforded by a virtual coin toss,
> > even a weighted one.  If the address selection must occur in the host, I
> > at least want a hook to influence the selection process (for srcaddr
> > *and* dstaddr).
>
> There have been many similar comments (from myself as well) asserting
> your statement.  It needs to be OS-independent as well, meaning that any
> device that speaks IPv6 should be able to follow my rules/hints for where
> traffic needs to go.  Any solution without this should be deemed
> unacceptable.

Agreed.

> Network mechanisms *need* to have visibility into alternate paths for a
> particular destination so that the path selection can be chosen via
> policy.

Exactly.  Note that the (srcaddr,dstaddr) tuple has to be chosen as a unit
for a particular path in order to avoid trouble with (e.g. anti-spoofing)
filtering.

On Wed, 18 Jun 2003, S Woodside wrote:
} I understand what you are saying, however, what about the opposite end,

Is asymmetry more of an issue in IPv6 than with IPv4?  (I'm really asking.
It might be.)

} is there a need to /always/ have management of path selection?

I'm arguing that there is always a need to have the /option/ of managing the
path selection.  The option might not always have to be used, but if we don't
build in the option now, it won't be there when needed.

________________________________________________________________________
Jay Ford, Network Engineering Group, Information Technology Services
University of Iowa, Iowa City, IA 52242
email: jay-ford@uiowa.edu, phone: 319-335-5555, fax: 319-335-2951