Subject: Re: [TE-wg] TE use in today's networks

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

    > From: Jennifer Rexford <jrex@research.att.com>

    >> If one *does* need dynamic response, it should be intuitively
    >> obvious that even if one can find a set of weights that produces
    >> the result one wants in the base case, then the resulting traffic
    >> pattern, once chance has taken some random set of elements out of
    >> service, is probably not going to be one that comes anywhere near
    >> meeting whatever goals one has.

    > experiments suggest that in practice the weights that are good in
    > the base case often work pretty well in the failure case; when they
    > don't, a change or two does the trick.

This may be related to a generic problem with min-path routing
architectures (i.e. those which pick src->dest paths which minimize a
fixed metric), which is that in many topologies, you see "hot links" which
want to be overutilized (and "cold links" which want to be underutilized).

The old ARPAnet had this problem to some degree, only there they had
load-based (actually delay-based) metrics in the routing which tended to
ameliorate it - when one link loaded up, traffic tended to divert around
that link. (Of course, then they had problems with oscillatory behaviour,
but that's a different problem!)

One obvious way to fix this (other than either varying the capacity of
links based on observed loads, or changing the topology so that it doesn't
have as many hot-spot problems) is to tweak the metrics on hot-spot links
up, to move traffic off them. So perhaps this part of what's going on.


    > What if you don't have a screwdriver and the hammer works pretty
    > well?

That's a perfectly appropriate attitude for operational people who have to
deal with what's available off the shelf.

However, I'd hope that when it comes to basic engineering, there's more of
a sense of "how do we do this right", not "how do we bludgeon existing
stuff into doing more or less what we want".

	Noel