On Tue, Sep 13, 2011 at 3:46 PM, Lianhu Wei <lianhu.wei.gmail.com> wrote:
> Hi everybody,
>
> I have installed AMBER 11 on a cluster, which consist os one head node and
> 50 cluster nodes. My OS is Red Hat enterprise linux. I used openMPI to
> compile amber parallel version. On head node and each of the cluster
> nodes,
> I can run msander.MPI on many threads (just on one node). But when I
> submit
> the job on multiple nodes, the many threads were still run on one node,
> instead of different nodes. I use PBSPro queue system. Here is my PBS
> script:
>
> #!/bin/csh
> #PBS -V
> #PBS -r n
> #PBS -N PAD2_94_Min
> #PBS -l select=8:ncpus=8
> #PBS -m abe
> #PBS -q verylong
> #
> set WRKDIR=/home/william/Work/PAD4/MD/KP94_PAD2
> cd $WRKDIR
>
> nohup mpirun -np 64 sander.MPI -O -i PAD2_KP94_CnsV_10ps.in \
> -o PAD2_KP94_CnsV_10ps.out \
> -p K94_PAD2.top \
> -c K94_PAD2_20K_rst \
> -x PAD2_KP94_CnsV_10ps.mdcrd \
> -v PAD2_KP94_CnsV_10ps.mdvel \
> -e PAD2_KP94_CnsV_10ps.mden \
> -r PAD2_KP94_CnsV_10ps.rst \
> -inf PAD2_KP94_CnsV_10ps.mdinfo
>
Don't use nohup in a PBS script. Also, I would suggest passing mpirun the
$PBS_NODEFILE as the hostfile/machinefile/whatever OpenMPI likes to call
it. That will tell mpirun where to lock the threads (rather than just
dumping them all on the same node).
HTH,
Jason
>
> Does anybody has ideas what went wrong?
>
> Thanks,
> William
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 13 2011 - 15:00:02 PDT
