[Pw_forum] Swapping

Gabriele Sclauzero gabriele.sclauzero at epfl.ch
Wed Oct 26 09:08:24 CEST 2011


Dear Payam,

   the solution might be simple: increase the number of nodes! The FFT-related quantities will be spread across more nodes, so that the memory per node will decrease (assuming that the number of pools stays the same).
It depends however on your system. If you have a very large one in hands (I mean, with a lot of electrons), you need to distribute also the matrix to be diagonalized (in cdiaghg). To achieve this, be sure to compile the scalapack enabled version and to use the -ndiag option properly. 
If you don't have more nodes, then I'm afraid you'll be forced to reduce the size of your system or PW cutoffs/number of bands. Also, using 'cg' style diagonalization will save some memory, but if this will be effective it depends on your system (about which you don't say a word...). If you need more help please put an effort to be more specific.

GS

Il giorno 25/ott/2011, alle ore 19.43, Payam Norouzzadeh ha scritto:

> Dear Quantum Espresso Users
> 
> I think the jobs run slow when they are swapping memory to disk.
> Is there anything I can do to distribute the memory across more nodes so they don’t have to swap?
> I want to reduce RAM usage in practice. How it is possible?
> 
> Thanks in advance.
> Payam Norouzzadeh
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum


§ Gabriele Sclauzero, EPFL SB ITP CSEA
   PH H2 462, Station 3, CH-1015 Lausanne







-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.democritos.it/pipermail/pw_forum/attachments/20111026/c911a3d1/attachment.htm 


More information about the Pw_forum mailing list