[ofa-general] RE: [PATCH] opensm: compute local geometry
rpearson at systemfabricworks.com
Sun Nov 30 08:56:36 PST 2008
I am looking at the earlier posts.
I had thought about this one before. All the cases where this algorithm
applies have low port counts. I can fix this by just not doing the
determinant if the port count is larger than the highest order polynomial in
the table since none of them will match.
From: Sasha Khapyorsky [mailto:sashak at voltaire.com]
Sent: Sunday, November 30, 2008 10:40 AM
To: Robert Pearson
Cc: general at lists.openfabrics.org
Subject: Re: [PATCH] opensm: compute local geometry
On 18:36 Sun 30 Nov , Sasha Khapyorsky wrote:
> On 10:59 Tue 11 Nov , Robert Pearson wrote:
> > Sasha,
> > Here is the fifth patch implementing the mesh analysis algorithm.
> > This patch implements
> > - routine to compute characteristics polynomial of a matrix
> > - routine to compute the local 'metric' around each switch
> I checked performance of determinant calculation - when switch has 8
> links it takes 11-12 seconds per switch, with 10 links - 2177 seconds.
Oops, sorry. The results above are for 10 and 12 links.
More information about the general