The MPLS-OPS Archive

[Antoine Bagula <bagula@cs.sun.ac.za>]

On Sat, 25 Sep 2004, Zartash Afzal Uzmi wrote:

> In a simulated environment, for 26 nodes, 10,000 random LSPs, one of my
> graduate student was able to place primaries as well as local backups within
> 9 seconds on a 2.something GHz wintel laptop in Java.

Please note that this processing speed also depends also on the path
computation algorithm used. It can take 9 seconds for a simple algorithm which
uses the Dijsktra algorithm with a specific cost metric while several hours may be required
to simulate the 10,000 random LSPs for an algorithm which involves several maxflow computations
(for example MIRA in full mesh).  The size of your fully meshed network has also a certain impact
on the processing speed.

> Computing only primaries will take perhaps 10-20% of this time. Throw in a C environment on
> a specially designed hardware, I would expect less than a second.
>
> > -----Original Message-----
> > From: mpls guy [mailto:mplsguy@gmail.com]
> > Sent: Saturday, September 25, 2004 1:17 AM
> > To: zartash@lums.edu.pk
> > Subject: Re: [MPLS-OPS]: Re: CSPF numbers...
> >
> >
> > I am just looking for some base line numbers  I know it would depend
> > upto the network size and constraints. It would not be that different
> > from the SPF numbers. For argument sake say there are 20 nodes and
> > they are fully meshed with priority and bandwidth attributes, I am
> > curious to findout what might be the CSPF values.
> >
> >
> >
> > On Sat, 25 Sep 2004 00:40:22 +0500, Zartash Afzal Uzmi
> > <zartash@lums.edu.pk> wrote:
> > > There is no fixed answer. The time to compute this is primarily
> > dependent
> > > upon the processor you throw in your LSR for doing this computation. It
> > > would also depend upon the constraints you specify, size of the
> > network, and
> > > to some extent how much pre-existing traffic you have on the
> > network. I am
> > > not even talking about the software efficiency...
> > >
> > > zartash
> > >
> > > > -----Original Message-----
> > > > From: mpls guy [mailto:mplsguy@gmail.com]
> > > > Sent: Friday, September 24, 2004 7:10 AM
> > > > To: mpls-ops@mplsrc.com
> > > > Subject: [MPLS-OPS]: Re: CSPF numbers...
> > > >
> > > >
> > > > what I mean is how much time dose it take for cspf to calculate when
> > > > you have 100, 200, 300 ingress lsp's. I am just looking for scalabilty
> > > > numbers or baseline info.
> > > >
> > > >
> > > > On Thu, 23 Sep 2004 17:23:31 -0400, mpls guy
> > <mplsguy@gmail.com> wrote:
> > > > > Does anyone have CSPF numbers matrix like accaptable cspf
> > numbers for
> > > > > 100, 200, 300 lsps etc...
> > > > > thanks
> > > > >
> > > >
> > > > -------
> > > > The MPLS-OPS Mailing List
> > > > Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> > > > Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
> > > >
> > >
> > >
> > >
> > > -------
> > > The MPLS-OPS Mailing List
> > > Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> > > Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
> > >
> >
>
> -------
> The MPLS-OPS Mailing List
> Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
>

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml