[Pw_forum] about output wavefunction, charge density and potential

J G Che jgche at fudan.edu.cn
Tue Apr 4 09:56:00 CEST 2006


However, now there are many abinit packages based plane-waves which don't use disk but RAM to store wavefunctions, meaning it could be at least as a option. Otherwise, such an often writing on disk not only wastes too much time, but also leads to NFS crashing, even for a small system. Each wavefunction files for a job with e.g. 50 atoms for 16 nodes is not more than 200M, it could be run in a PC-Cluster with a standard configuration of 1G/CPU, no problem. However, two these jobs running leads to NFS very slow or crashing, since every ten-seconds it needs to write 6G data on disk. 

JG
  ----- Original Message ----- 
  From: Stefano Baroni 
  To: pw_forum at pwscf.org 
  Sent: Tuesday, April 04, 2006 1:43 PM
  Subject: Re: [Pw_forum] about output wavefunction, charge density and potential




  On Apr 4, 2006, at 3:02 AM, J G Che wrote:


    Dear Paolo:


    Thanks for you replying. 


    Do you mean that pwscf uses disk to store wavefunctions? 


    I think it is not necessary to store wavefunction on disk, since now the RAM is very sheap. 


  well, this statement is, strictly speaking, not much meaningful until one specifies 


  1) how greedy one is and 
  2) how large is one's wallet ...


  SB




    Che




    ----- Original Message ----- 
    From: "Paolo Giannozzi" <giannozz at nest.sns.it>
    To: <pw_forum at pwscf.org>
    Sent: Tuesday, April 04, 2006 2:55 AM
    Subject: Re: [Pw_forum] about output wavefunction, charge density and potential




      On Saturday 01 April 2006 05:10, J G Che wrote:


        I find that in each iterative, wavefunction, potential and charge density
        will be saved on disk. Who knows how to output these once only when 
        the self-consistent is reached. Thanks!


      - if you have more than one k-point per processor, you have to save
        wavefunctions to disk. It would be possible to store them into memory 
        but it would require extensive changes to the code (and a lot of memory)
      - if you have just one k-point and do not specify disk_io='high' the
        code will write wavefunctions only at convergence
      - the charge density and potential are written at each scf step.
        This is a behavior that could be changed with simple modifications
        in the code. These arrays however are usually much smaller than
        the wavefunctions.
      - unless you specify disk_io='high', auxiliary arrays used for
        self-consistency are not written to disk but stored
      - other files are typically small


      Paolo
      -- 
      Paolo Giannozzi             e-mail:  giannozz at nest.sns.it
      Scuola Normale Superiore    Phone:   +39/050-509876, Fax:-563513 
      Piazza dei Cavalieri 7      I-56126 Pisa, Italy


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


  ---
  Stefano Baroni - SISSA  &  DEMOCRITOS National Simulation Center - Trieste
  [+39] 040 3787 406 (tel) -528 (fax) / stefanobaroni (skype)


  Please, if possible, don't  send me MS Word or PowerPoint attachments
  Why? See:  http://www.gnu.org/philosophy/no-word-attachments.html






-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/attachments/20060404/ce399668/attachment.htm 


More information about the Pw_forum mailing list