[Pw_forum] Trying to find reasonable parallelization parameters

[Axel Kohlmeyer]

On Thu, 22 Jan 2009, J. J. Ramsey wrote:

JJR> The cluster I'm running on uses Infiniband for MPI communications, 
JJR> and the queueing software by default limits me to 1280 processor 
JJR> cores for a single job, 1.8 GB per core, and a wall time of 96 
JJR> hours. As warned about in the user guide, I'm running into problems 
JJR> with running out of memory when using pw.x, so I need to find 
JJR> appropriate values for npool, ntg, and ndiag. (I don't have an NEB 
JJR> calculation, so I don't thing that the # of images is an issue.) 
JJR> I'm a bit unsure about sane values of these, since as far as I can 
JJR> tell, my simulation isn't a "very large" job but isn't exactly 
JJR> small, either: 213 atoms, 727 electrons, and I'm using 4x4x1 
JJR> k-point sampling. (It's a slab calculation.)

JJR> How many cores, roughly, is reasonable for a simulation like this, 
JJR> and what's a reasonable way to divide it up into pools, ortho 
JJR> groups, etc.?

jj,

very simple. -npools can be maximally the actual number of k-points that 
you are using, and that should be exploited to the maximum. a second 
option to pay attention to, is the disk_io flag. setting this to 'low' 
can have a very positive effect on scaling at the expense of additional 
memory use. for the rest you have to make a sequence of benchmarks to 
see how far you can push it. there is no way to predict this. too many 
factors beyond the contol of the source code have some impact. for some 
of the advanced options, the code tries to measure its efficiency at the 
beginning of a run.

cheers,
   axel.


JJR> 
JJR> 
JJR> 
JJR>       
JJR> _______________________________________________
JJR> Pw_forum mailing list
JJR> Pw_forum at pwscf.org
JJR> http://www.democritos.it/mailman/listinfo/pw_forum
JJR> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.