[gmx-users] parallel gromacs

Scott Delinger scott.delinger at ualberta.ca
Tue Jun 11 22:06:35 CEST 2002

Thanks to Choon Peng and Erik Lindahl for their help on my first 
benchmark question. I seem to get the correct run times for Athlon 
1533 performance within 5%. Now for parallel....

I've got lam installed (RHL 7.3 distro-provided RPM). I've got 
gromacs installed (3.1.3-1 SRPM, rpm --rebuild, rpm -i). I've got 
MPICH installed as the "preferred" MPI on the system, which 
interfered with trying to install GROMACS from source with MPI 
enabled. I've installed gromacs-mpi-3.1.3-1.i386.rpm (likely not 
optimised for my hardware). I've unpacked the benchmarks, and tried 
running d.vallin (grompp -np 2 -shuffle -sort), but stdout shows that 
all of the atoms fall in one CPU. Fair enough, but I see Athlon 
(dual) on the benchmark page and wonder how that was accomplished.

I then move on to d.dppc, which looks like it's meant for parallel 
work. If I run grompp -np 2 -shuffle -sort and then mdrun -np 2 (from 
the original gromacs SRPM), stdout from the mdrun complains that 
GROMACS was compiled without parallel capabilities (but the stdout 
from grompp shows that the job is prepared to run across two 
processors evenly). Fair enough, --enable-mpi likely wasn't in that 
source RPM. If I run grompp -np 2 -shuffle -sort and then mdrun_mpi 
-np 2, stdout from mdrun_mpi complains that it was expecting an input 
file with np = 1 and found np = 2. (!) If I run grompp -shuffle -sort 
and then mdrun_mpi, it runs on a single processor with no performance 
difference from the standard runs I did yesterday and this morning.

I'm obviously not seeing something correctly. I'm not sure why 
mdrun_mpi wants a -np 1 version of the input files. I'm wanting it to 
run across both processors on the one box. I'll worry about running 
it more widely on the full set of machines 
(http://www.chem.ualberta.ca/~scott/) once I've solved my easier (?) 

Thanks for suffering through all that verbiage, and thanks in advance 
for any help you can offer.

Scott L. Delinger, Ph.D.		IT Administrator
Department of Chemistry
University of Alberta
Edmonton, Alberta, Canada  T6G 2G2
scott.delinger at ualberta.ca

More information about the gromacs.org_gmx-users mailing list