Recompile amber11 in parallel, being *sure* to use the same MPI that you're
using the mpirun from. The only times I've seen this happen is when Amber
is installed with the mpif90/mpicc from one MPI implementation (i.e.
OpenMPI), and mpirun/mpiexec is used from another implementation (i.e.
mpich2).
What's actually happening is 8, single-processor sander.MPI processes are
spawned, rather than 1, 8 processor job in which all 8 procs communicate.
HTH,
Jason
On Fri, Jul 22, 2011 at 2:36 PM, M Kang <medichem.gmail.com> wrote:
> Dear all,
>
> I submit my sander.MPI (amber11) job with qsub using the following script:
> ----
> #!/bin/sh
>
> INPUT=sander.in
>
> cat $PBS_NODEFILE
>
> /opt/amber11-ambertools-1.4/
> bin/lamboot -d -v $PBS_NODEFILE
>
> echo "mpirun" 1>&2
>
> /opt/amber11-ambertools-1.4/bin/mpirun -np 8
> /home_changlab/croberts/local/amber11/bin/sander.MPI -O `cat sa
> nder.in`
>
> echo "lamclean" 1>&2
>
> /opt/amber11-ambertools-1.4/bin/lamclean
>
> echo "lamhalt" 1>&2
> /opt/amber11-ambertools-1.4/bin/lamhalt
> -----
>
> Then I run "top" command, which shows 8 processors are running sander.MPI
> about 100%.
>
> But the output says it is running on 1 nodes.
> ---
> | Atom division among processors:
> | 0 3128
> | Running AMBER/MPI version on 1 nodes
> ---
>
> And actually it is more than 3 times slower than a sander.MPI (amber10) job
> with 4 nodes in the other machine.
> ---
> | Atom division among processors:
> | 0 797 1564 2361 3128
> | Running AMBER/MPI version on 4 nodes
> ---
>
> It doesn't show any error message. How can I fix this problem?
>
> *Waiting for some suggestion in this regards.
>
> Myung*
> _______________________________________________
> 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 Fri Jul 22 2011 - 14:00:04 PDT
