[gmx-users] [Re]: Preliminary report on benchmark on Opteron Cluster Infiniband on current CVS version

[Luca Ferraro]

> Date: Tue, 17 Apr 2007 19:08:02 +0200 From: Erik Lindahl
> <lindahl at cbr.su.se> Subject: Re: [gmx-users] Preliminary report on
> benchmark on Opteron Cluster Infiniband on current CVS version To:
> Discussion list for GROMACS users <gmx-users at gromacs.org> Message-ID:
> <242D227B-350E-4172-AA9D-A4CF0427BD47 at cbr.su.se> Content-Type:
> text/plain; charset=US-ASCII; format=flowed Hi, On Apr 17, 2007, at
> 6:26 PM, Luca Ferraro wrote:
>>> /usr/bin/time mpiexec -n ${proc} mdrun_mpi -d 1 ${proc} 1
> 
> For a start, the domain decomposition flag is "-dd"  :-)
> 
> Cheers,
> 
> Erik
> 
> 

<homer_mode> doh!! </homer_mode> :-P

.. well, a nice "warning" from mdrun complaining
"what did you mean with that -foo option?!"
would be really appreciated in the next release ;-)

lferraro at poseidon:~/benches/d.lzm>mdrun -foo
[SNIP] (with no warnings)
Reading file topol.tpr, VERSION 3.3.99_development_20070215 (single 
precision)
starting mdrun 'LYSOZYME (E.C.3.2.1.17) in water'
5000 steps,     20.0 ps.

OK, I will run these benches again and post (I hope correct this time) 
results. I also ask for some clarification:

1) Does the -dd vector (x,y,z) really reflect a spatial splitting of the 
simulation BOX along the x,y,z axes? What happens to bonded molecules 
crossing the splitting? Is there any documentation or notes about 
technictian details on this new parallel implementation? (I mean beyond 
the source code).

2) -rdd   real      0  The minimum distance for DD communication
what exactly means? (can you give me any example?)

Again, thank you in advance for any support.

luca



-- 
+------------------------------------------+
| Luca Ferraro
| Gruppo Scienze dei Materiali
| CASPUR (Consorzio per le Applicazioni di
| Supercalcolo Per Università e Ricerca)
| Via dei Tizii, 6b - 00185 ROMA
| Tel. +39-06-44486717
| Fax: +39-06-4957083
| cell: +39-339-7879898
| Email: luca.ferraro at caspur.it
| Web: http://www.caspur.it
+------------------------------------------+