[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [bmwg] Thoughts ondraft-vapiwala-bmwg-rsvpte-convergence-motivation-01.txt
Hello Adrian,
It is great that this work is interesting you, and we would be very
happy to present this in the MPLS working group as well.
As we are working towards streamlining the scope of this effort, the
primary focus of this work is to elaborate on the key performance and
scalability metrics that one need to consider evaluating the nodes to be
used in a MPLS-TE network. In this process, we consider convergence as
one of the key performance and scalability metrics. The convergence
within the context of this effort is defined as how long does it takes
to signal RSVP-TE based label switched paths following a network
convergence event. We will also try to bring this out more evidently in
the document to eliminate the ambiguity.
Furthermore, we do not consider IGP convergence under this umbrella
since it is being benchmarked separately. We can definitely make sure
that the TE extensions are either covered under this part of under IGP
convergence.
For section 3, we will consider the RSVP Hellos and timers
For Section 3.1, we will clarify that
For Section 5, these are recommended scalability values that one may
consider to measure convergence. We could always limit this to the 1
LSP, but in order to come with the behavior or impact of scale on
convergence this range is necessary. One could always argue if this
recommended range is the optimal one, and we are open to suggestions.
For Section 6, Since there is a possibility of performance being
sacrificed due to security being enabled, we should caution the reader.
We will consider addressing this
Also thanks for the Nits, we will address these in the future version.
Thanks in advance.
Authors
-----Original Message-----
From: Adrian Farrel [mailto:adrian at olddog.co.uk]
Sent: Friday, September 28, 2007 2:45 PM
To: bmwg at ietf.org
Subject: [bmwg] Thoughts
ondraft-vapiwala-bmwg-rsvpte-convergence-motivation-01.txt
Hi,
I have been reading
draft-vapiwala-bmwg-rsvpte-convergence-motivation-01.txt
and have some thoughts.
First, I think it might be wise to socialise your work with the MPLS
working
group who (presumably) have some interest in this subject.
It may be obvious in the context of the BMWG, but I struggled to
understand what "convergence" applied to. Normally, I think, convergence
applies to the
process of the information stored on a particular node (or all nodes in
the
network) reaching stability that reflects the distributed state of the
network. Thus IGP convergence. In the context of MPLS-TE I might have
expected you to be discussing the convergence of the TE extensions to
the IGP, but your I-D is very focused on RSVP-TE, and the explanation of
"convergence" in your context doesn't really come out until the
performance cases in section 4.
Section 3 should include consideration of RSVP Hellos.
Section 3.1 It is very good to mention refresh reduction, but without
mentioning soft state refresh, the context is lost.
I am unsure what the objective of section 5 is. I would caution you
against allowing anything that could be construed as marketing
information from creeping in. The problem of defining an apple before
trying to compare it with another apple seems not to have been
addressed.
I think section 6 needs some more thought. One of the things you suggest
measuring is the performance impact of applying RSVP security. At the
least,
you should caution against security being sacrificed for performance.
You might like to reference (or at least glance at
draft-yasukawa-mpls-scaling-analysis-04.txt).
Nits
---
Filename header
/-convergece-/-convergence-/
---
Abstract
s/compliments/complements/
---
Section 1
s/providers achieve/providers to achieve/ s/in proportional/in
proportion/ s/This high-availability/Because of these high-availability/
Cheers,
Adrian
_______________________________________________
bmwg mailing list
bmwg at ietf.org
https://www1.ietf.org/mailman/listinfo/bmwg
_______________________________________________
bmwg mailing list
bmwg at ietf.org
https://www1.ietf.org/mailman/listinfo/bmwg