[mpich-discuss] Suppressed Output

[SULLIVAN David (AREVA)]

Eduardo,
I read that in the manual as well, but I think it may be referring to the negative value of tasks to disable load balance and error tolerance. Using OMP through the tasks variable is actually much easier to use than going through the mpiexec process. Its a quick test to see...of course you could try the mcnp group as well. I know the mcnpx forum is closed to beta users, but the mcnp5 forum has users of both versions and a surprising number are involved in parallel processing as well.

Regards,

Dave


-----Original Message-----
From: mpich-discuss-bounces at mcs.anl.gov on behalf of Padilla, Eduardo Antonio
Sent: Wed 9/8/2010 5:59 PM
To: mpich-discuss at mcs.anl.gov
Subject: Re: [mpich-discuss] Suppressed Output
 
Thanks Dave - I think the tasks option is only available with PVM versions, which I do not have built. I'd like to avoid learning how to do this (the reason for me simply downloading the beta MPI version and installing mpich2). Hopefully, there's a simple way to correct the cpu time error.
-Eduardo

From: mpich-discuss-bounces at mcs.anl.gov [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of SULLIVAN David (AREVA)
Sent: Wednesday, September 08, 2010 3:50 PM
To: mpich-discuss at mcs.anl.gov
Subject: Re: [mpich-discuss] Suppressed Output

Eduardo,

I am more familiar with mcnp5 than x but I think this is still applicable- For using a single machine it is often better to use the mcnp input variable "tasks 8" (for your situation), rather than going through mpiexec. So it would look like:

mcnpx tasks 8 i=filename.i o=filename.o

If the executable has been built for threading..if it bombs with the error "multiprocessing is unavailable in this compilation", you will need to rebuild it for OMP (if you have the source files). Hope this is helpful-

Dave

________________________________
From: mpich-discuss-bounces at mcs.anl.gov [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of Padilla, Eduardo Antonio
Sent: Wednesday, September 08, 2010 5:13 PM
To: mpich-discuss at mcs.anl.gov
Subject: [mpich-discuss] Suppressed Output
All,
I recently installed the current version of MPICH2 1.2.1p1 on a 32-bit XP Pro machine with 8 cores (the machine is 64-bit but the OS is 32). I'm trying to run MCNPX threading all 8 cores. My command line is as follows:
mpiexec -localonly 8 mcnpx i=filename...

This ran great exactly twice. However, now when I run it I see MCNPX start up normally, create the initial files and then:
Dynamic storage = 0 words, 0 bytes. Cp0 = 0.00
Master starting 7 by 1 subtasks 09/03/10 (time)
Master sending static commons...
Master sending dynamic commons...
Master completed initialization broadcasts.
Rendezvous at nps = 1000 # of microtasks = 7 (date time - this is MCNPX language here)
Rendezvous at nps =***************** of microtasks = 7 (date time - this is MCNPX talk but confusing to me)

At this point, all 8 cores are running full bore in task manager, but the process gets hung up. The output files are never written or updated (data dumps don't occur). When I run based on a set number of particle histories (not based on cpu time), the process seems to work fine. I assume a simple setting got changed somewhere?
Regards,
Eduardo Padilla
Sandia National Labs


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/ms-tnef
Size: 4478 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20100908/2d94635e/attachment.bin>