[Developers] Add OpenMP statistics to PUGH

John Shalf jshalf at lbl.gov
Fri Oct 23 11:43:24 CDT 2009


On Oct 23, 2009, at 7:43 AM, Frank Loeffler wrote:
> Hi,
>
> On Fri, Oct 23, 2009 at 07:39:56AM -0700, John Shalf wrote:
>> it is a very bad thing of processes migrate due to NUMA effects.
>> Data gets pinned to the memory controller that is closest to the
>> processor that first touched it.  Things get very inefficient if the
>> process migrates to a distant processor.
>
> Yes, but you do not always have different memory controllers  
> attached to
> different cores.

Correct,
it is usually on a per-socket basis (except when AMD and Intel jam two  
chip dies onto a single socket, which they do on occasion).

>> You might also take a look at JumpShot (inside of MPE) to see how
>> they do their instrumentation.
>> ftp://ftp.mcs.anl.gov/pub/mpi/mpe/mpe2-1.1.1.tar.gz
>> It works on BlueGene as well, and profiles any OpenMP implementation
>> that is built on top of pthreads (as most of them are).
>
> thanks for the that pointer.
>
> Frank
>
> _______________________________________________
> Developers mailing list
> Developers at cactuscode.org
> http://www.cactuscode.org/mailman/listinfo/developers



More information about the Developers mailing list