The MPLS WG Archive

[Kireeti Kompella <kireeti@juniper.net>]

On Fri, 14 May 2004, Adrian Farrel wrote:

> > Discussion points:
> > a) does 3209 (as implemented today) imply hard or soft preemption?
>
> What do you mean "does 3209 (as implemented today) imply..."

Semantics, semantics.  To state this more precisely, albeit somewhat
longwindedly, I should have said, "If you have an implementation,
insofar as such implementation is one of RSVP-TE (hereafter called
3209), and such implementation incorporates the notion of preemption,
does the said implementation interpret 3209 to mean that the receipt
of a new LSP request that contends for resources already occupied by
existing established LSPs of lower priority than the aforementioned
new LSP request would immediately and irrevocably cause the resources
thus contended for to be unhesistatingly released in favour of the new
LSP and the LSPs using those resources to be unceremoniously aborted
(hereafter called hard preemption), or would cause the new LSP request
to held up, or indeed established in conjunction with the pre-existing
LSPs, awaiting a more graceful release of the resources and
termination of the heretofore established LSPs (soft preemption)?"

(and of course, a corresponding question for deployments).

Personally, I prefer French. :-)

> The draft may or may not imply something.

I apologize for my loose use of language.  I agree with you: a
discussion of what 3209 implies is non-productive.

> Why can we not accept that clarification is needed, and move on by
> writing that clarification?

Accepted.  The questions I asked were intended to set the direction of
such a clarification.

> If you want an anlysis of what has been implemented, I refer you to
> draft-farrel-mpls-preemption-interim-00.txt which I wrote in November at the request of
> the WG chairs and which contains the responses from 10 implementors.

I did read that; seems rather inconclusive (5 hard, 4 soft, 1 neither).
It would be good to get rough consensus on what the clarification doc
should say.

> > b) should the baseline preemption policy for 3473 follow whatever we
> >    agree that 3209 implies?
>
> Erm, I take it you are asking the question "is GMPLS a superset of
> MPLS?", in which case the answer is yes.
>
> However, if you are asking whether 3473 will should be
> clarified/changed in line with whatever clarification/change is made
> to 3209, and whether that will impact deployed implementations, you
> are asking the question on the wrong list.

Agreed.  Being of a work-conserving nature (hereafter called lazy), I
wanted to do both in one shot; however, if we can make progress on
question (a) on this list, I will follow up question (b) on the CCAMP
mailing list.

Kireeti.
-------