The MPLS WG Archive

["Brijesh Kumar" <bkumar@ennovatenetworks.com>]

Dmitri Krioukov writes:
> -----Original Message-----
> From: owner-mpls@UU.NET [mailto:owner-mpls@UU.NET]On Behalf Of Dmitri
> Krioukov
> i'd like to hear any points of view of the members
> of this list on motivation #1 (te) accompanied
> with some far going criticism to this reference:
> http://smg.ulb.ac.be/Preprints/Fortz99_29.html
> (it was brought up to discussion on nanog
> mailing list recently).
>
> i understand that the major assumption in this paper
> (about the demand matrix availability) is too
> brave, but this paper is a major indication
> that the te-related research beyond the ever
> lasting vc reinvention approach has not been
> exhausted yet. clearly, you wouldn't need any
> external signaling mechanism if all benefits
> te brings you (congestion avoidance, qos, etc)
> may be achieved by means of some (rather
> futuristic at this point) routing protocol.

That is a very good case specific paper (also a good theoretical paper),
however the solution is  dependent on the demand matrix (demand between two
nodes s and t) D(s,t,T) remaining constant for a substantial period of time
(T). The authors recognize that the solution will be lot more harder to
implement when this assumption is not true. The overheads of Routing
protocols being directly dependent on the frequency  at which network states
must be updated in routers, makes the goal of supporting time variant
traffic demands without proportionally increasing routing and related
signaling overheads very hard, if not impossible, to achieve. On the other
hand, as you rightly said, the difficulty of designing routing protocols
that can naturally accommodate TE should not be precluded as an approach.
But that approach better match MPLS TE functionality and have less overheads
to be acceptable. Kind of hard task..

At this point, MPLS remains the best long term solution for TE.

> at the very general level, the packet switched
> networks proved to provide much higher level of
> adaptation to the new requirements. since the
> packet switched networks are theoretically
> more complex than circuit switched ones, the way
> to adapt to a new requirement (te) may not be obvious
> until some significant research effort is done. this
> does not mean that we have to avoid this effort since
> the easy way may not lead anywhere...

Good point. I am sure even MPLS design can benefit from those efforts.

--brijesh
Ennovate Networks Inc.