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

RE: update of JSA for inter-domain diverse path & some simulation results



Hello All,

Just to add to what Fabio said ...

First, the draft is now officially on-line, and is available
at
http://www.ietf.org/internet-drafts/draft-dachille-diverse-inter-region-path
-setup-01.txt

Indeed, the simulation results are helping to provide more insight
into the working of these schemes, and in developing better
intuition about them.

We would encourage CCAMPers to take a look at the revised draft,
and provide their inputs.

We intend to discuss the revised draft and the simulation
results at DC, and look forward to your feedback.

-Vishal

> -----Original Message-----
> From: ricciato [mailto:ricciato@coritel.it]
> Sent: Wednesday, October 27, 2004 3:52 AM
> To: ccamp@ops.ietf.org
> Cc: marco@infocom.uniroma1.it; Ugo Monaco; Daniele Alì; Alessio
> D'Achille; Vishal Sharma
> Subject: update of JSA for inter-domain diverse path & some simulation
> results
>
>
> Hi all,
>
> please note that we have have updated the draft on the JSA  method for
> computing diverse paths in inter-domain.
> We have also run some initial simulations on realistic topologies,
> comparing the 1) JSA approach (with ARO) with 2) the method based on
> RRO+XRO (called ISPA in the draft), and 3) the global optimum based on
> complete information.
> We considered a few realistic topologies (details in the draft)
> In summary the results show that :
>
> - if a pair of diverse path exist, the JSA/ARO almost always find them
> at the first shot (it failed in 2-3 cases over thousands of
> simulation run)
> - there are some topologies in which the ISPA/XRO is trapped at the
> first shot (and should therefore revert to cranckback), it happens
> approx. in 10% of cases
> - when a pair of path is found by all the three methods, there is no
> substantial difference in the overall cost between JSA/ARO and optimum
> (we used min-hop metric), with a slight worse performance of ISPA/XRO
>
> Please note that the performance of PCE should be the same as the global
> optimum (since it is computed assuming full global information and
> centralized computation).
>
> Based on these preliminary results, we believe that the JSA/ARO approach
> should not be considered a competitor of PCE (which is provenly "the
> optimum"), but rather a secondary solution which achieves the same
> performances of the optimal method  "almost surely" in the practical
> cases  (still, it is possible to draw cases in which it fails, while
> PCE/optimum finds his way, but these cases seem to be not found in the
> realistic topologies we have investigated).
> On the other hand the JSA/ARO  involves much  less information and
> signaling overhead...
>
>
> In addition to the IETF site, the current version of the draft can be
> found at the following url. It includes quick  description of the
> simualtions and summary of preliminary results. It would be nice to have
> a short slot for presenting them at Washington DC, perhaps ....
>
> We wellcome very much any feedback from the community that helps us in
> refining the simulation scenario (new topologies, suggestions for
> setting the scenario, etc.). Please do not hesitate to send you comment
> (particularly the criticisms, which are the most helpful ones for
> improving our work ...)
>
> ciao
> fabio
>
> PS
> the url for the draft is:
>
> http://www.metanoia-inc.com/Publications/draft-dachille-diverse-in
ter-region-path-setup-01.txt