[mpich-discuss] profiling (gprof) MPICH2 manager program

[William Pearson]

Thanks.  That is exactly what I needed.

Bill Pearson


Begin forwarded message:

> From: Darius Buntinas <buntinas at mcs.anl.gov>
> Date: May 18, 2010 11:17:43 AM EDT
> To: mpich-discuss at mcs.anl.gov
> Cc: William Pearson <wrp at virginia.edu>
> Subject: Re: [mpich-discuss] profiling (gprof) MPICH2 manager program
> 
> What's probably happening is that processes are clobbering each other's gmon.out files.  If you set GMON_OUT_PREFIX to something it will use that filename and append the pid of the process.  That should reduce the chance of the files getting clobbered (you could still have processes with the same pid on different nodes, but that's not likely).
> 
> E.g.:
>  GMON_OUT_PREFIX=mygmon.out mpiexec -n 10 cpi
> 
> -d
> 
> On 05/18/2010 09:55 AM, William Pearson wrote:
>> 
>> I am interested in looking for bottlenecks to an MPICH2 program that prevent efficient scaling after 10 - 16 worker programs.
>> 
>> I was hoping to use gprof to do this, but the gmon.out file that I get is overwhelmed by the worker program times -- I'm not seeing the relevant timings for the manager program (which sends data and collects results).
>> 
>> How can I tell gprof to focus only on the first instance of the program that is running as the manager?
>> 
>> Bill Pearson
>> 
>> _______________________________________________
>> mpich-discuss mailing list
>> mpich-discuss at mcs.anl.gov
>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20100518/5520d93f/attachment.htm>