[Pw_forum] parallel processing

[somayeh fotohi]

dear all 
I have two problems in quantum espresso.


I work a simple structure of graphene.(60 atom)
I run my program in cluster with 32 core (4 node) and use  below command for running:


export PATH=/opt/espresso/openmpi-1.4.2-install/bin:/opt/intel/composer_xe_2011_sp1.11.339/bin/intel64:$PATH
export OMP_NUM_THREADS=1
/opt/espresso/openmpi-1.4.2-install/bin/mpirun -hostfile .nodelist60 -np 32 /opt/espresso/espresso-5.0/bin/pw.x -npool 8 < 60.scf.in > 60.scf.out


output file was reported:

 General routines
     calbec       :   1832.29s CPU   2261.50s WALL (   20586 calls)
     fft          :     21.02s CPU    105.36s WALL (     130 calls)
     ffts         :      1.83s CPU      4.52s WALL (      29 calls)
     fftw         :   1026.00s CPU   2618.48s WALL (   33010 calls)
     interpolate  :      6.29s CPU     13.35s WALL (      29 calls)
     davcio       :      0.01s CPU  23906.83s WALL (      88 calls)
 
     Parallel routines
     fft_scatter  :    295.81s CPU   1939.74s WALL (   33169 calls)
     EXX routines
 
    PWSCF        :     2h 8m CPU       10h48m WALL

as you seen , difference between CPU and WALL is very much and  run of program is finished after 12 hours approximately.
what can i do for reduce WALL time?


another problem is :
when i use 32 cores or 16 cores,the speed dose not change considerably. I do not know the parallel of cores is done  badly or must change the parameter for improving run time and performance.


can any one help me?

thanks for advance.
Best Regards
somayeh fotoohi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20120829/4dec2e31/attachment.html>