[molpro-user] Molpro 2006.1 mpirun

[Manhui Wang]

Dear Joseph Lane,
     For your problem, I just quote a probably temporary solution on
SGI Altix system. Hopefully it helps.

===============================================================

Comments to your request #7693 by aaron.mcdonough:
 ......

For your problem:

I found that in utilities/molpro.c there is some logic to check for the
PBS_NODEFILE environment variable. If it is set, which it is for cherax
in batch mode (as we use PBS), then molpro will set a '-machinefile
$PBS_NODEFILE' argument for mpirun. Unfortunately, the SGI/Altix mpirun
doesn't have this option. You can workaround this by unsetting the
PBS_NODEFILE in your script:


#!/bin/bash
#PBS -l walltime=2:0:0
#PBS -l ncpus=2
#PBS -l vmem=40gb

cd $PBS_O_WORKDIR
echo $PBS_NODEFILE && cat $PBS_NODEFILE

unset PBS_NODEFILE

make MOLPRO_OPTIONS="-v -n2" test

...

===================================================================================

Best regards

Manhui Wang


Joseph Lane wrote:
> I have just compiled molpro 2006.1 on our SGI Altix system. The test
> jobs all ran fine, and my initial benchmarks look good. However, I can
> no longer submit jobs using our Torque/Maui batch queue system. The
> equivalent scripts with molpro 2002.6 work fine, but it seems the way
> mpirun is used is different.
> 
> It seems that molpro 2006.1 is incorrectly guessing the number of cpus
> from my submission script, and uses a different syntax for mpirun that
> causes the following error
> 
> /usr/bin/mpirun -machinefile
> /usr/spool/PBS/aux//2253.ou019132.otago.ac.nz -np 1 /usr/bin/dplace -s1
> /home/qchem/molpro2006.1/bin/molprop_2006_1_i8_ia64_mpi.exe
> MPI: bad process count
> 
> The number of processors always defaults to 1, irrespective of the
> number of cpus requested.
> 
> 
> From what I can tell, SGI's version of mpirun does not allow the
> "-machinefile" option.
> 
> 
> How can I disable this 'automatic' guessing of ncpus so that it behaves
> more similar to molpro 2002.6
> 
> 
> 
> Any help would be greatly appreciated
> 
> 
> 
> Kind regards
> 
> 
> 
> 
> Jo
> 
> 
> 
> 
> 
> --Joseph Lane
> Department of Chemistry
> University of Otago
> P.O. Box 56
> Dunedin, New Zealand
> ph +64 3 479 5378
> email : jlane at alkali.otago.ac.nz
> 

-- 
-----------
Manhui  Wang
School of Chemistry, Cardiff University,
Main Building, Park Place,
Cardiff CF10 3AT, UK
Telephone: +44 (0)29208 76637
E-mail: wangm9 at cardiff.ac.uk, manhuiwang at gmail.com