[gmx-developers] Problem with qmmm and mpi combo (4.0.5)

[Christian Seifert]

Hi!

I never managed to run GMX with MPI when using QMMM. I tried it with different 
versions (also GMX 3). As Berk wrote, most of the time (in my case more than 
95%) the QM program has to calculate. I guess that running the MM part in QMMM 
with multiple CPU will be even slower compared to MM part of QMMM on a single 
CPU (because of communication overhead).
So my decision was:
I do not compile GMX_qmmm with mpi.

Greets,
Christian

Am Freitag, 20. November 2009 um 16:40 schrieb Ake Sandgren:
Hi!

When building gromacs 4.0.5 for qmmm and mpi something goes wrong
(unless its in the input data of course :-)

For the input data I'm using (from a user of ours) i have the following
problem:
In qmmm.c update_QMMMrec the gmx_sumi in the parallel excecution part
have inconsistent values in md->nr, i.e. not the same on process 0 and
process 1. (Causing a problem in the Allreduce)

Also fr->QMMMlist_sr.nri is different between processes.

What might be the cause of this? Input data error or bug in the code?

I've been trying to backtrace this but it's now friday afternoon and i'm
giving up :-(

-- 
Ake Sandgren, HPC2N, Umea University, S-90187 Umea, Sweden
Internet: ake at hpc2n.umu.se   Phone: +46 90 7866134 Fax: +46 90 7866126
Mobile: +46 70 7716134 WWW: http://www.hpc2n.umu.se

-- 
gmx-developers mailing list
gmx-developers at gromacs.org
http://lists.gromacs.org/mailman/listinfo/gmx-developers
Please don't post (un)subscribe requests to the list. Use the 
www interface or send it to gmx-developers-request at gromacs.org.

M.Sc. Christian Seifert
Department of Biophysics
University of Bochum
ND 04/67
44780 Bochum
Germany
Tel: +49 (0)234 32 28363
Fax: +49 (0)234 32 14626
E-Mail: cseifert at bph.rub.de
Web: http://www.bph.rub.de