[molpro-user] run molpro in parallel with sge

[Reuti]

Hi,

Am 30.11.2007 um 17:58 schrieb Javier Díaz Montes:

> I am trying to run molpro in parallel. That is, I want to execute a  
> molpro job in two nodes of a cluster using mpi. However I obtain an  
> error when I execute my batch file. Someone could I help me?
> I have installed molpro with this rpm molpro-mpp-2006.1-102.p4.rpm.
>
>
> This is the error:
> ---------------------------------------------------------------------- 
> ----
> mpirun was unable to determine how many processes to launch for the
> following process:
>
>     /home/programs/molpro/molpro-mpp-Linux-i686-i4-2006.1/ 
> molprop_2006_1_i4_p4_tcgmsg.exe
>
> You must specify how many processes to launch, via the -np
> argument.
> ---------------------------------------------------------------------- 
> ----
>
> This is my batch file
>
> #!/bin/bash
> #$ -S /bin/bash
> #$ -V
> #$ -cwd
> #$ -pe mpich 2
>
> /home/programs/molpro/bin/molpro is2qrtbt.com

we use:

molprop -n $NSLOTS --mpirun-machinefile $TMPDIR/machines < infile >  
outfile

> And this is the molpro execution file:
>
> #!/bin/sh
> TMPDIR=/state/partition2/$JOB_ID

$TMPDIR is already set by SGE (i.e., it will create and delete a  
directory dedicated for each job on all involved nodes). Only thing  
to note is, that you should setup your PE that it is only attached to  
one queue, as the queuename will be part of the directory, and it  
should be the same on all nodes.

For this to work, you will need a so called Tight Integration of your  
parallel application. With a proper set up PE, the in SGE defined  
start_proc_args will create a link to a rsh-wrapper. Was GA compiled  
with rsh or ssh? You used the MPI template in your SGE setup?

> mkdir $TMPDIR
> MOLPRO_CONFIG="\
> -G 8000000 \
> -d$TMPDIR \
> -I$TMPDIR \
> -W$SGE_O_WORKDIR \
> -m7272727 \
> -K131072 \
> -B64 \
> --tcgmsg \

If you compiled GA to use TCGMSG over MPI, this should read: --mpi \

> -lmpirun \
> --environment MOLPRO_COMPILER=ifort9.1 \
> --environment MOLPRO_MXMBLK=0064 \
> --environment MOLPRO_MXMBLN=0064 \
> --environment MOLPRO_CACHE=0012288 \
> --environment MOLPRO_UNROLL=2 \
> --environment MOLPRO_MINDGM=5000 \
> --environment MOLPRO_MINDGC=5000 \
> --environment MOLPRO_MINDGR=5000 \
> --environment MOLPRO_MINDGL=5000 \
> --environment MOLPRO_MINDGV=5000 \
> --environment MOLPRO_FLOPMXM=001961 \
> --environment MOLPRO_FLOPDGM=001961 \
> --environment MOLPRO_FLOPMXV=001302 \
> --environment MOLPRO_FLOPDGV=001302 \
> -x /home/programs/molpro/molpro-mpp-Linux-i686-i4-2006.1/ 
> molprop_2006_1_i4_p4_tcgmsg.exe \

We use TCGMSG over MPI and get molprop_2006_1_i4_p4_mpi.exe

> -L /home/programs/molpro/molpro-mpp-Linux-i686-i4-2006.1/ \
> ";
> export MOLPRO_CONFIG
> exec /home/programs/molpro/molpro-mpp-Linux-i686-i4-2006.1/molpro $*

Additonal hints if you use MPICH1 in SGE:

http://gridengine.sunsource.net/howto/mpich-integration.html

and their mailing lists:

http://gridengine.sunsource.net/maillist.html

especially "users".

-- Reuti