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

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

On Mon, 16 Jun 2003, J. Noel Chiappa wrote:
> Load balancing is, fundamentally, a routing issue. You're selecting among one
> of several alternate paths.
>
> This is not a routing group, really - the basic consensus seems to be that we
> can't efficiently support lots-and-lots of multi-homing in the routing, and
> we're going to do it by having multiple addresses, and selecting among them.
>
> So you're asking in the wrong place.
>
>     > (or at least it would be easy to set up even for a naove user.)
>
> You can do some load-balancing by ugly hacks like chosing addresses.
>
> However, I'm afraid your stated goal (load-balancing for idiots) is in
> fundamental conflict with one tool that people are currently using to do some
> load-balancing, i.e. address selection - address selection is tricky and
> doesn't work in all cases, and is therefore inappropriate for a "works for
> idiots" mechanism.

The problem I see is that (srcaddr,dstaddr) selection by a multihomed IPv6
host in the context of provider-based addressing can preempt the routing.
That means that the loading will be determined by clueless edge hosts, which
is about the worst place for such decisions to be made.  This feels like a
fundamental problem with the current direction of IPv6 deployment for which I
haven't yet seen a solution (except possibly a beefed up version of the NAROS
idea, which would effectively remove the address selection responsibility
from the host).

________________________________________________________________________
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