[gmx-users] Running GROMACS in cluster with SGE

[Yasser Almeida Hernández]

Hi all,

I am trying to run a simulation on a cluster with SGE queue system with 
this script:

*******
#!/bin/bash
#$ -cwd
#$ -pe parallel 16 #number of cores
#$ -q main.q  #name of the queue
#$ -q main.q at xbes23.akbester.zz #one can choose particular node
#$ -e stderr.log   # redirections for stdout,err
#$ -o stdout.log

export I_MPI_FABRICS=shm:ofa

mpirun mdrun -v -deffnm sim1

*******

For some reason I get a very poor performance around 0.8 ns/day in a 
node with 16 cores, compared with a workstation with 8 cores, where I 
get 2 ns/day.

I am running GROMACS 4.5.5 installed from the Ubuntu repository in each 
node. Could this be the reason for the poor performance?
Should I compile GROMACS from source code on the cluster (cluster 
optimized)?

Thanks in advance

Yasser

-- 
Yasser Almeida Hernández
PhD student
Institute of Biochemistry and Molecular Biology
Department of Chemistry
University of Hamburg
Martin-Luther-King-Platz 6
20146 Hamburg
Germany
+49 40 42838 2845
yasser.almeida.hernandez at chemie.uni-hamburg.de
office: Grindelallee 117, room 250c