[gmx-users] MPI tips

[Erik Lindahl]

Hi,

As explained in the FAQ ( http://www.gromacs.org/faq ), you only need  
a dedicated MPI version of mdrun.

On many platforms it works fine to just build the entire package with  
MPI enabled, but since some queue systems disable you from  
dynamically linking with MPI libraries (e.g. if it requires dedicated  
access to special hardware) it is possible to build everything  
without MPI and then just build mdrun with MPI support.

As for running things with MPI, you don't even need the -np flag  
anymore - it's autodetected. Just use -np X as an argument to grompp  
in order to  split your topology for X processors.

How you actually start parallel jobs in MPI environments isn't  
standardized, so you'll have to check the documentation of your MPI  
implementation.

For LAM-MPI it's a matter of

1. lamboot -v <hostfile>
2. mpirun -np X mdrun_mpi <standard mdrun arguments>

Cheers,

Erik


On Jan 30, 2006, at 4:59 PM, David Mathog wrote:

> I'm trying to figure out how to use MPI with gromacs.  A few examples
> would go a long, long way.  Where are the MPI examples????
>
> Which programs support MPI???
>
> When I built gromacs two versions of (all) the programs ended up in
> .../gromacs/bin, a "regular" one and ones with an _mpi suffix. The two
> versions for each program are about the same size, with the _mpi one
> being either 32 bytes bigger or smaller (go figure).  That could just
> be compiler noise.
>
> The "regular" ones work (at least those run from gmxdemo/demo.sh
> which is my translation to bash of the original csh script.)
>
> Now I'm trying to work through demo_mpi.  The demo_mpi that
> gromacs built is IDENTICAL to demo, and even with lam-mpi running
> (yes, I tested that MPI works with other programs) it only runs
> on the head node.   The man page says that mdrun has an -np
> parameter if built with MPI.  I couldn't find any other useful
> info on whether or not there were actually MPI versions of the
> other programs, _mpi suffix or not.
>
> I'm stuck at the very first program which is pdb2gmx.  The man page
> for that doesn't indicate any -np parameter for it. I tried running
> pdb2gmx_mpi and it just ran on the head node with no activity on the
> compute nodes.
>
> Thanks,
>
> David Mathog
> mathog at caltech.edu
> Manager, Sequence Analysis Facility, Biology Division, Caltech
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
>

-----------------------------------------------------------
Erik Lindahl  <lindahl at sbc.su.se>     Backup address:  
<erik.lindahl at gmail.com>
Assistant Professor, Computational Structural Biology
Stockholm Bioinformatics Center, Stockholm University, SE 106 91  
Stockholm
Phone: +46 8 5537 8564     Fax: +46 8 5537 8214