[gmx-users] gmx_alltoall

[Carsten Kutzner]

Anna Marabotti wrote:
> Dear GMX-developers (in particular dear Carsten Kutzner),
> to overcome problems in making parallel runs with GROMACS on a Linux cluster with Gigabit Ethernet
> interconnection, I downloaded the package gmx_all-to-all to speed up the processes. In the
> instructions, however, this package is indicated for GROMACS version 3.3.1, whereas in our cluster
> we installed the 3.3.2 version. Is gmx_alltoall compatible also with this version?

Hi Anna,
in principle the patch should also work with 3.3.2. I have just applied 
the patch to 3.3.2, and then taken a quick look at the patched code 
which looks good, so no apparent problems here.

>  Another question: if we will upgrade to GROMACS 4 when it will be 
released, are you expecting that
> (nearly) all problems in parallelizazion will be solved or this package will be still
> useful/required/necessary?

The CVS version still uses the MPI_Alltoall routine for the parallel FFT 
within PME, so you could run into the same problems on Ethernet.

What you could do is to move over to a recent version of OpenMPI which 
comes with optimized collectives. They provide an MPI_Alltoall routine 
which switches between several algorithms depending on message size and 
number of processors. I think this is turned on by default, so all you 
need to do is install OpenMPI, recompile Gromacs on top of that and you 
should not have problems with all-to-all congestion any more.

Carsten


> Many thanks in advance and best regards
> Anna Marabotti
> 
> ______________________________________________
> Anna Marabotti, Ph.D.
> Laboratorio di Bioinformatica e Biologia Computazionale
> Istituto di Scienze dell'Alimentazione, CNR
> Via Roma 52 A/C
> 83100 Avellino (Italy)
> Tel: +39 0825 299651
> Fax: +39 0825 781585
> Skype: annam1972
> E-mail: amarabotti at isa.cnr.it
> Web page: http://bioinformatica.isa.cnr.it/anna.htm
> ____________________________________________________
> "If you think you are too small to make a difference, try sleeping with a mosquito"
> 
> _______________________________________________
> 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

-- 
Dr. Carsten Kutzner
Max Planck Institute for Biophysical Chemistry
Theoretical and Computational Biophysics Department
Am Fassberg 11
37077 Goettingen, Germany
Tel. +49-551-2012313, Fax: +49-551-2012302
http://www.mpibpc.mpg.de/research/dep/grubmueller/
http://www.gwdg.de/~ckutzne