[gmx-users] No scale up beyond 4 processors for 240000 atom system

Diego Enry diego.enry at gmail.com
Tue Oct 9 20:40:31 CEST 2007


Low cost tip:
Ask your cluster administrator if it is possible apply channel bonding
to the Gigabit interfaces. You need two network switches for that to
be efficient (a cut-through switch may also help). It increases
network bandwidth by 70%. It also helps to use Cat6 cables.

You may try this MPICH compilation:
./configure --with-device=ch3:nemesis --enable-fast --disable-cxx
--enable-error-checking=no CFLAGS=-O3 FFLAGS=-O3

There is patch to run PME faster over ethernet... maybe that paper
from Dr. Carsten that Dr. Christian suggested.

High cost tip:
Migrate to Infiniband.


Anyone tried GotoBLAS ? Does it work well with GMX ?


On 10/9/07, Christian Burisch <burisch at bph.rub.de> wrote:
> Berk Hess schrieb:
>
> Hi all,
>
> > So this is 4 cores sharing one ethernet connection?
>
> perhaps the two Gigabit NICs were bundled somehow. But I guess this
> doesn't work out-of-the-box&plug'n'play. And latency and not bandwidth
> may be limiting in this case.
>
> > With such a setup you will never get good scaling.
> > You need something like an Infiniband network.
>
> Or check:
>
> Carsten Kutzner, David van der Spoel, Martin Fechner, Erik Lindahl, Udo
> W. Schmitt, Bert L. de Groot and Helmut Grubmuller: Speeding up parallel
> GROMACS on high-latency networks J. Comp. Chem 28 pp. 2075-2084 (2007)
>
> Haven't tried it yet but sounds good!
>
> Regards
>
> Christian
>
> --
> Dr. Christian Burisch
> Lehrstuhl für Biophysik
> PG Theoretische Biophysik
> Ruhr-Universität Bochum
> D-44780 Bochum
> Raum ND04/67
> Fon: +49 234 32 28363
> Fax: +49 234 32 14626
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>


-- 
Diego Enry B. Gomes
Laboratório de Modelagem e Dinamica Molecular
Universidade Federal do Rio de Janeiro - Brasil.



More information about the gromacs.org_gmx-users mailing list