[Pw_forum] MareNostrum

[Jose C. Conesa]

Dear Emmanuel,
Your make.sys file and mine are identical except that:

a) I have swapped the lines defining F90FLAGS and FFLAGS so that the 
former, which uses FFLAGS, comes afterwards (otherwise the said use is 
useless...). I normally do so as I never understood why the standard 
make.sys does it otherwise.
b) I am using openmpi rather than intelmpi (I have #BSUB -a openmpi in 
my script). This affects to the definition of SCALAPACK_LIBS, which in 
my case contains -lmkl_blacs_openmpi_lp64 instead of  
-lmkl_blacs_intelmpi_lp64.
c) My make.sys contains a line saying ELPA_LIBS_SWITCH = disabled. It's 
not clear to me in this moment why it is so.

Please note that, if you find your problem only when running neb.x, I 
cannot guarantee that my compilation is free from the same error since I 
do not use that program.
Best,
José Carlos

El 09/03/2013 15:44, Emmanuel Clouet escribió:
> Dear Jose,
>
> I have concluded that there was a memory leak problem because my jobs 
> are swapping. Tanks to the help of MareNostrum support, we have seen 
> that the swapping does not occur at the beginning of the calculation 
> but after some time. In a NEB calculation for instance, the first NEB 
> iteration runs perfectly well, but there is a strong increase of 
> memory when the program starts the second NEB iteration. I think this 
> is a problem with my compilation, because the same kind of calculation 
> run perfectly well on other HPC platforms.
> I attached to this Email a copy of a typical make.sys file I used. 
> This corresponds to the 5.0 version of QE compiled with Intel compiler 
> (v13.0.1) using MKL (13.0.1), FFTW3 (3.3) and intel MPI (4.1.0.024) 
> libraries. I also tried with QE 5.0.3 version, with internal FFT 
> instead of external FFTW3 library, as well as with open MPI (1.5.4) 
> instead of Intel MPI, without any success.
> Is it possible that you send me your make.sys file so I can test it 
> and see if it solves my problem.
> Best regards,
>
> Emmanuel Clouet
>
> On 08/03/2013 23:31, Jose C. Conesa wrote:
>> Hi Emmanuel,
>> I did compile it. I use only pw.x, so I can tell nothing about the 
>> behaviour of neb.x, although it was generated. I am also not sure if 
>> I would be able to spot a memory leak problem. How did you observe it?
>> Regards,
>> José Carlos Conesa
>>
>> El 08/03/2013 21:30, Emmanuel Clouet escribió:
>>> Dear all,
>>>
>>> Does anybody manage to compile QE (the pw and neb parts) on 
>>> MareNostrum III (Barcelona Supercomputing Center)? I am encountering 
>>> some memory leak problems with my compiled version and I am looking 
>>> for somebody who is using QE on this platform and can help me with 
>>> the compilation.
>>> Best regards,
>>>
>>> Emmanuel 
>
>
> -- 
> Emmanuel CLOUET
> SRMP - Bat. 520
> CEA - Saclay
> 91191 Gif sur Yvette cedex
> tel. +33 (0)1 69 08 66 63
> http://emmanuel.clouet.free.fr
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum


-- 
José C. Conesa
Instituto de Catálisis y Petroleoquímica, CSIC
Marie Curie 2, Cantoblanco
28049 Madrid - Spain
Tel. +34-915854766

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20130309/a8e97380/attachment.html>