[Pw_forum] K-point parallelization

Axel Kohlmeyer akohlmey at vitae.cmm.upenn.edu
Sun Nov 27 21:38:16 CET 2005


On Sat, 26 Nov 2005, Sergey Lisenkov wrote:

SL> 
SL>  Dear PWscf authors,
SL> 

hi sergey,

SL>   I am wondering does k-point parallelization work very well on any
SL> platform? I have noticed that on our cluster (IBM Blade PPC970 -
SL> power4+) PW code works slowly with k-point parallelization in
SL> respect to without k-point parallelization. Does it depend on the
SL> platform?

you have to make sure, that you have a fast and preferably 
local scratch area available and use it. if you are using 
a networked scratch, your i/o load with increase immensely,
since each k-point 'pool' will write to the scratch concurrently.
on a networked scratch, especially over gigabit ethernet, this
will kill the performance.

axel.

p.s.: any news on the ru-potentials?

SL> 
SL>   Thanks,
SL>     Sergey
SL> _______________________________________________
SL> Pw_forum mailing list
SL> Pw_forum at pwscf.org
SL> http://www.democritos.it/mailman/listinfo/pw_forum
SL> 

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







More information about the Pw_forum mailing list