Re: [AMBER] PARALLEL PERFORMANCE OF AMBER12

From: Ross Walker <ross.rosswalker.co.uk>
Date: Fri, 16 Aug 2013 09:52:53 -0700

Hi Uday,

My guess would be that this openmpi has not been configured to use the IB
interface and so is sending things over ethernet. Try the OSU micro
benchmarks to make sure your latencies and bandwidths are good.

http://mvapich.cse.ohio-state.edu/benchmarks/

All the best
Ross



On 8/16/13 5:40 AM, "Uday Midya" <umidya.iitkgp.gmail.com> wrote:

>Hi Ross
>Thanks for your reply. I am using openMPI.
>
>which mpirun ---->/opt/openmpi/bin/mpirun
>
>which mpif90 ---->/opt/openmpi/bin/mpif90
>
>echo $PATH
>----->/opt/openmpi/bin:...................................................
>...................
>
>Does this MPI correct.
>
>I used the following command for the simulation. I did not use the queuing
>system.
>
>mpirun --hostfile lam.hosts sander.MPI -O -i ComplexTestHeat.in -o
>ComplexTestHeat.out -p ComplexTest.prmtop -c ComplexTestMinimz.rst -r
>ComplexTestHeat.rst -x ComplexTestHeat.mdcrd
>
>lam.hosts file is
>
>compute-0-0 cpu=8
>compute-0-1 cpu=8
>
>Am I doing the right things?
>
>Thanks
>Uday
>_______________________________________________
>AMBER mailing list
>AMBER.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Aug 16 2013 - 10:00:02 PDT
Custom Search