[gmx-developers] Parallelized g_rdf

[Anton Feenstra]

On 09/06/10 22:53, Nikos Bentenitis wrote:
> Hello all,
>
> I would like to request for two features to be included in future
> versions of g_rdf.
>
> 1. Would it be possible to parallelize g_rdf? Because I am simulating
> 20 thousand molecules and averaging center-of-mass rdfs for 2.5 ns,
> g_rdf takes approximately 20 hours on a single CPU.
> 2. I am simulating solute A in solvent B, and I need to calculate
> three radial distribution functions: g_AA, g_BB and g_AB. With the
> current implementation of g_rdf, I need to run the program twice: once
> to
> calculate g_AA, and g_AB and once to calculate g_BB. Because I am
> simulating 20 thousand molecules, each run takes about 20 hours on a
> single core, for a total of approximately 40 hours on a single core.
> Would it be possible to have g_rdf produce all three rdfs for each
> run?

The simple answer is, yes, two times.

The real question is, of course, who will do it. If you have some 
proficiency in programming, you should be able to at least implement 
option 2.

By the way, needing two runs of g_rdf already gives you a 
parallelization since you can run them independently. Possibly, although 
it is hard to be sure without trying, you can split you 20 thousand 
molecules over (many) CPUs to the same effect?


-- 
Groetjes,

Anton
  _____________ _______________________________________________________
|             |                                                       |
|  _   _  ___,| K. Anton Feenstra                                     |
| / \ / \'| | | IBIVU/Bioinformatics - Free University  Amsterdam     |
|(   |   )| | | De Boelelaan 1083A - 1081 HV Amsterdam - Netherlands  |
| \_/ \_/ | | | Tel +31 20 59 87783 - Fax +31 20 59 87653 - Room P136 |
|             | Feenstra at few.vu.nl - www.few.vu.nl/~feenstra/         |
|             | "Pump Up the Volume Along With the Tempo" (Jazzy Jeff)|
|_____________|_______________________________________________________|