[Gllug] Re: CPU exerciser?

[John Hearns]

On Tue, 2004-10-26 at 09:34, Jan Kokoska wrote:

Talking about CPU burning,
its fun to watch the power being drawn as we run CPU burn tests
The amperage draw on a big cluster can be large :-)

> And more specific, oh well, I would see if there is an
> instruction-monitoring plugin for Valgrind, which is a x86
> debugger/profiler (and indeed has a number of plugins for quite specific
> purposes). I remember the (very apt) developers were not too worried
> about implementing every single MMX/SSE/whatever instruction and rather
> waited for people to post their complaints into list.. the new one
> always appeared immediately in CVS head. You would need to valgrind your
> HPL test program and find what it does on that place.. then try and
> reproduce it. Would end up in coding, I am afraid I don't know anything
> cooked already. Overkill IMHO.
Strange you should mention Valgrind.
I've been interested in it for a few years, since there were some memory
leaks in high energy physics code.

I looked at it again last week for another reason - there is no x86-64
version yet,
The mailing list says that they are working on it.
Look forward to using it when it comes out.




> It seems to me like you just need a good excuse to thow it back at your
> distributor, right? Saying it freezes with HPL should do, they can see
> themselves. Just swap it it through a few boxes to make sure it's the
> cpu.. I am sure you have *lots* of those ;)
The CPU will be replaced. We have 43 other processors in that cluster
hammering along fine. Its a 22 node cluster with Myrinet, intended for
computational chemistry. Those types just soak up all the CPU available,
and Opteron is doing very well in that market.
I'm just trying to do a bit of diagnostics, and maybe learn something,
before the thing is swapped out.





-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug