[Pw_forum] НА: Re: problem in MPI running of QE (16 processors)

Huiqun Zhou hqzhou at nju.edu.cn
Tue Mar 8 05:48:57 CET 2011


Alexander,

According to your reply to my message, you actually applied 64 CPU cores 
(16 nodes, 4 cores per node), this should have no problem unless the policy
of using your cluster prohibited it. Once upon a time, we had such a policy
on our cluster: an job occupies at most 32 CPU cores, otherwise put it into
sequential queue.

Maybe, you should ask your administrator whether there is such a policy ...

zhou huiqun
@earth sciences, nanjing university, china
 
  ----- Original Message ----- 
  From: Alexander G. Kvashnin 
  To: PWSCF Forum 
  Sent: Tuesday, March 08, 2011 12:24 AM
  Subject: Re: [Pw_forum] НА: Re: problem in MPI running of QE (16 processors)


  Dear all



  I tried to use full paths, but it didn't give positive results. It wrote an error message  


  application called MPI_Abort(MPI_COMM_WORLD, 0) - process 0




  On 7 March 2011 10:30, Alexander Kvashnin <agkvashnin at gmail.com> wrote:

    Thanks, I tried to use "<" instead of "-in" it also didn't work.
    OK,I will try to use full paths for input and output, and answer about result.

    ----- Исходное сообщение -----
    От: Omololu Akin-Ojo <prayerz.omo at gmail.com>
    Отправлено: 7 марта 2011 г. 9:56
    Кому: PWSCF Forum <pw_forum at pwscf.org>
    Тема: Re: [Pw_forum] НА: Re: problem in MPI running of QE (16 processors)

    Try to see if specifying the full paths help.
    E.g., try something like:

    mpiexec /home/MyDir/bin/pw.x -in  /scratch/MyDir/graph.inp >
    /scratch/MyDir/graph.out

    (where /home/MyDir/bin is the full path to your pw.x and
    /scratch/MyDir/graph.inp is the full path to your output ....)

    ( I see you use "-in" instead of "<" to indicate the input. I don't
    know too much but _perhaps_ you could also _try_ using "<" instead of
    "-in") .

    o.

    On Mon, Mar 7, 2011 at 7:31 AM, Alexander Kvashnin <agkvashnin at gmail.com> wrote:
    > Yes, I wrote
    >
    > #PBS -l nodes=16:ppn=4
    >
    > And in userguide of MIPT-60 wrote,that mpiexec must choose number of
    > processors automatically, that's why I didn't write anything else
    >
    >
    > ________________________________
    > От: Huiqun Zhou <hqzhou at nju.edu.cn>
    > Отправлено: 7 марта 2011 г. 7:52
    > Кому: PWSCF Forum <pw_forum at pwscf.org>
    > Тема: Re: [Pw_forum] problem in MPI running of QE (16 processors)
    >
    > How did you apply number of node, procs per node in your job
    > script?
    >
    > #PBS -l nodes=?:ppn=?
    >
    > zhou huiqun
    > @earth sciences, nanjing university, china
    >
    >
    > ----- Original Message -----
    > From: Alexander G. Kvashnin
    > To: PWSCF Forum
    > Sent: Saturday, March 05, 2011 2:53 AM
    > Subject: Re: [Pw_forum] problem in MPI running of QE (16 processors)
    > I create PBS task on supercomputer MIPT-60 where I write
    >
    > mpiexec ../../espresso-4.2.1/bin/pw.x -in graph.inp > output.opt
    > all other

    [Включен не весь текст исходного сообщения]



  -- 

  Sincerely yours
  Alexander G. Kvashnin
  --------------------------------------------------------------------------------------------------------------------------------
  Student
  Moscow Institute of Physics and Technology          http://mipt.ru/
  141700, Institutsky lane 9, Dolgoprudny, Moscow Region, Russia



  Junior research scientist

  Technological Institute for Superhard 
  and Novel Carbon Materials                                   http://www.ntcstm.troitsk.ru/
  142190, Central'naya St. 7a, Troitsk, Moscow Region, Russia
  ================================================================




------------------------------------------------------------------------------


  _______________________________________________
  Pw_forum mailing list
  Pw_forum at pwscf.org
  http://www.democritos.it/mailman/listinfo/pw_forum
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20110308/05fec3d7/attachment.html>


More information about the users mailing list