[gmx-users] Parallel message passing

[Nathan Moore]

> On Wed, 2005-06-08 at 15:25 -0500, Nathan Moore wrote:
>> I have a parallel version of gromacs running on a commercial ppc machine
>> with many nodes available.  Whn I look at parallel scaling (only the
>> villin benchmark so far) I see linear speedup only up to about 10
>> processors.  Is villin (~9000 atoms) too small to show proper speedup?
>> I
>> mean to look at the parallel benchmark in a few days.
> Scaling Villin to 10 processors is great. I haven't heard of that
> anywhere else.
>>
>> In reading through the manual it sounds like GROMACS only passes
>> messages
>> to it's nearest neighbors (node_i only talks to node_{i+1} and
>> node_{i-1})
>>  Is this really how things are set up?  I have a full-featured MPI
>> network
>> available that I'd like to utilize fully - is an alternate
>> communications
>> protocol set up?
> The PME algorithm uses full all-to-all communication. However in general
> you want to reduce communication to a minimum (by using clever
> algorithms). What kind of network do you have?

A very fast one.  Processors are relatively slow (~700MHz ppc32), nodes
are connected on a torus, distributed memory, 256 or 512MB/processor node
- no virtual memory.  I compile with xlC, to be practically useful GROMACS
has to scale to >=512 processors...

>
>>
>> NT Moore
>> _______________________________________________
>> 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.
> --
> David.
> ________________________________________________________________________
> David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group,
> Dept. of Cell and Molecular Biology, Uppsala University.
> Husargatan 3, Box 596,          75124 Uppsala, Sweden
> phone:  46 18 471 4205          fax: 46 18 511 755
> spoel at xray.bmc.uu.se    spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>
> _______________________________________________
> 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.
>