[Pw_forum] Trying to find reasonable parallelization parameters

[Axel Kohlmeyer]

On Thu, 22 Jan 2009, Timothy Mason wrote:

TM> I have a follow up question on this issue.
TM> 
TM> if I get the following message at the beginning of the output file
TM> 
TM> "Iterative solution of the eigenvalue problem.
TM> Too few procs for parallel algorith we need 4 procs per pool
TM> a serial algorithm will be used"
TM> 
TM> does this suggest that if I assign 2 procs per pool then I'm wasting a processor?

tim,
only a benchmark can tell for certain. in most cases the k-point 
parallelization is by far the most efficient. thus you should 
first try to use as many pools as possible. even if you are not
using a parallel diagonalization algorithm, you are still using
the g-space parallelization and distribute the wavefunction
and density (see summary at the beginning), which coincides with
parallelization of several time consuming operations. 

HTH,
  axel.


TM> 
TM> Tim Mason
TM> University of Missouri St. Louis
TM> _______________________________________________
TM> Pw_forum mailing list
TM> Pw_forum at pwscf.org
TM> http://www.democritos.it/mailman/listinfo/pw_forum
TM> 

-- 
=======================================================================
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.