[gmx-users] help on mdrun

[Mark Abraham]

On 9/07/2011 3:37 AM, raghuvir at bcpindia.org wrote:
> Hi
>
>
> I have a problem related to submitting a mdrun job on cluster.  I 
> tried to ask help or gromacs and rocks users-group.
>
> My machine specs.
> Cluster of Intel Xeon processors:QC with Rocks Cluster.  8 processors 
> (16 threads)
>
> When I submit mdrun -deffnm umbrella0 -pf pullf-umbrella0.xvg -px 
> pullx-umbrella0.xvg on the localhost, the job takes note of 16 threads 
> and the job completes in a day.  But when I try submitting it to the 
> cluster using following qsub script the job takes ~16 days
>
> -----------------------------------------------------------------------
> #!/bin/bash
> #
> #$ -cwd
> #$ -S /bin/bash
> #
> #$ -N umbrella0
> #$ -e umbrella0.errout
> #$ -o umbrella0.out
> #
> #$ -pe mpi 16
>
> echo -n "Running on: "
>
> /opt/openmpi/bin/mpirun -np 16 -machinefile /home/raghuvir/machines 
> /share/apps/gromacs-4.5.3/bin/mdrunmpi -deffnm umbrella0 -pf 
> pullf-umbrella0.xvg -px pullx-umbrella0.xvg
>
> echo "Done."
> ---------------------------------------

That's fine as far as GROMACS is concerned, so long as mdrunmpi really 
has been compiled with MPI. You can get what diagnostic information it 
knows about the MPI setup from the very top of the .log file. Otherwise, 
you'll have troubleshoot your use of MPI and your batch queue system.

Mark