[Beowulf] [gorelsky at stanford.edu: CCL:dual-core Opteron 275performance]

[Joe Landman]

Hi Mikhail and Jacob:

  We have done quite a bit of testing with dual core on a number of other
chemistry and informatics apps at customer sites, and written a white paper
about it.  The summary (punch-line) that we saw for Amber8 and GAMESS was that
dual CPU dual core results were effectively the same as quad CPU single cores.
 There are some cases where we expected memory contention to be an issue,
though we did not see this in our measurements.

  I am sure it is possible to find codes and test cases which will fill the
bandwidth for the memory controller.  The usual suspects (Amber8, GAMESS,
BLAST, HMMer) did not.

Joe

On Wed, 06 Jul 2005 21:59:33 +0400, Mikhail Kuzminsky wrote
> In message from Jakob Oestergaard <jakob at unthought.net> (Wed, 6 Jul 
> 2005 16:43:00 +0200):
> >On Wed, Jul 06, 2005 at 04:12:25PM +0200, Alan Louis Scheinine wrote:
> >...
> >> My main point is, as I wrote, "your milage may vary."  I've heard 
> >>from 
> >> various
> >> people that "everybody is going to dual-core".  I simply want to 
> >>emphasize 
> >> that
> >> the dual-core choice is not for everybody.  In particular, I looked 
> >>at 
> >> profiling
> >
> >Guys, the results are very interesting - it would be very interesting 
> >to
> >know, too, which kernels you have been running on.
> >
> >For some time there have frequently been fixes to the NUMA code in
> >general, and there's definitely been some changes for handling 
> >dual-core
> >processors in the opteron case.
> >
> >If, and I don't know if this is the case, the kernels you have tested
> >with do not properly understand the topology of your NUMA setup in
> >dual-core (two processors (cores) per node (memory controller)),
> >performance should definitely be expected to degrade noticably.
> >
> >I'm pretty sure at least 2.6.12 had NUMA/dual-core fixes in it. 
> >Anyway,
> >if you're running older 2.6 kernels it would probably be interesting 
> >to
> >try a more up-to-date kernel.
> I'll try to test dual cores Opteron w/2.4 kernel and 2.6 kernel,
> but if I'm correct, SuSE 9.3 (we plan to use) has a bit more old 2.6 
> :-(
> BTW, may be there some difference in tests performance between 2.4 
> and more old 2.6 kernels ?
> 
> Yours
> Mikhail
> 
> >
> >Thanks for the results and observations - it's a good read :)
> >
> >-- 
> >
> > / jakob
> >
> >_______________________________________________
> >Beowulf mailing list, Beowulf at beowulf.org
> >To change your subscription (digest mode or unsubscribe) visit 
> >http://www.beowulf.org/mailman/listinfo/beowulf
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf


--
Scalable Informatics LLC
http://www.scalableinformatics.com
phone: +1 734 786 8423