[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Last call comments: draft-ietf-tewg-interarea-mpls-te-req-01.txt



Hi,

At 09:21 PM 6/13/2004 +0100, Adrian Farrel wrote:
Hi,

Contrary to appearances, this discussion may be making progress.

JP said,

> > And I do not see why specifying the signalling method would be a problem
> > at all. If a SP requires a contiguous LSP to avoid its LSP to be
> > stitched along its path and potentially reoptimized without any HE
> > control, specifying that the LSP must be contiguous (or not) is indeed
> > the right method.

There are three issues there.

1. Is it a problem to specify the desired/required signaling method?

There are two answers.
a. There is no technical problem with providing this function.
b. There are plenty of fears about the consequences of providing
    this function. Dimitri and Vishal have listed some.

2. Can a HE reasonably require control of functions such as
    re-optimization?

The answer to this would appear to be "yes". But note that this requirement surely applies
to all signaling methods. It happens to be the case that the option is meaningless in
contiguous signaling because the LSP cannot be optimized downstream of the HE (using
existing make-before-break techniques - but who knows what the future holds?).


So here is a good example of a functional requirement ("re-optimization only under
head-end control") that the head-end should be able to signal. You should make sure this
is clear in section 7.9.


3. Is there a reason to require contiguous signaling rather than some other form?

I still don't see one.

Well, you may not want a downstream SP to use stitching and potentially "hide" what happens downstream in term of reroute, failure, reoptimization, ...


Please understand that I am *NOT* saying that SPs have a perfect right to control what
signaling is used within their network. This is usually achieved through the joint
measures of procurement and configuration. An SP would be justifiably vexed to discover
that a cluster of LSRs within their network had autonomously switched LSP setup requests
to operate in CR-LDP.
However, it is not true to say that when an ingress LSR sends an RSVP-TE Path message it
is requiring that that signaling technique be used all the way to the egress.


The issue clearly gets fuzzy when the LSP traverses part of the network that is out of the
originating SP's direct control (i.e. another AS). But here we are talking only about
areas.

But as you know the solutions will apply to both inter-area and inter-AS.


Cheers

JP.


Cheers,
Adrian