[QE-users] QE Parallelization for Phonon Dispersion calculations
Kiran Yadav
kiranyadav0816 at gmail.com
Tue Nov 24 10:39:19 CET 2020
Dear Lorenzo,
Is there any parallelization method so that I can make images such that 20
dyn files will get distributed unequally on the equal number of processors
or dyn files will get distributed equally on unequal number of processors ?
Because in my case as per my observation dyn1-10 are not taking too much
time, most of the time has been spent on dyn11-dyn20 generation. So, If I
could run dyn 1-10 on one image and distribute remaining dyn11-20 on 3
images CPU walltime can be decreased. Is it possible to do something like
that?
I tried parallelization on 256 processors (#PBS -l select=16:ncpus=16) for
10hrs using the following parallelization command
time -p mpirun -np $PBS_NTASKS ph.x -ni 4 -nk 4 -nt 4 -nd 16 -input ph.in >
ph.out
in this case 20dyn files got distributed in 4 images i.e 5 dyn on one image
( I guess). And each image runs on 256/4=64 processors, means my 64*2=128
processors are of no use after creating dyn 1-10. So, according to me If
could run dyn 1-10 on one image and can divide remaining 10 dyn files (dyn
11-20) on 3-4 images all processors will finish almost together.
Kindly suggest on the same.
Thanks & Regards,
Kiran Yadav
Research Scholar
Electronic Materials Laboratory (TX-200G)
Dept. of Materials Science & Engineering
Indian Institute of Technology, Delhi
On Tue, Nov 24, 2020 at 2:57 PM Kiran Yadav <kiranyadav0816 at gmail.com>
wrote:
> Dear Lorenzo,
> Is there any parallelization method so that I can make images such that 20
> dyn files will get distributed unequally on the equal number of processors
> or dyn files will get distributed equally on unequal number of processors ?
>
> Because in my case as per my observation dyn1-10 are not taking too much
> time, most of the time has been spent on dyn11-dyn20 generation. So, If I
> could run dyn 1-10 on one image and distribute remaining dyn11-20 on 3
> images CPU walltime can be decreased. Is it possible to do something like
> that?
>
> I tried parallelization on 256 processors (#PBS -l select=16:ncpus=16) for
> 10hrs using the following parallelization command
> time -p mpirun -np $PBS_NTASKS ph.x -ni 4 -nk 4 -nt 4 -nd 16 -input ph.in >
> ph.out
> in this case 20dyn files got distributed in 4 images i.e 5 dyn on one
> image. 256/4=
>
> Thanks & Regards,
> Kiran Yadav
> Research Scholar
> Electronic Materials Laboratory (TX-200G)
> Dept. of Materials Science & Engineering
> Indian Institute of Technology, Delhi
>
>
>
>
> On Tue, Nov 24, 2020 at 1:19 PM Lorenzo Paulatto <paulatz at gmail.com>
> wrote:
>
>>
>> > I have been trying to calculate Phonon dispersion with 6*6*6 nq grid.
>> > It generates dyn0+20 other dynamical matrices, but the time taken by
>> > each dynamical matrix file is different for completion. I ran these
>> > phonon dispersion calculations using parallelization, but couldn't
>> > optimize correctly.
>>
>> This is normal, because different q-points have different symmetries,
>> the code has to use different numbers of k-points.
>>
>>
>> regards
>>
>>
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20201124/cbe2ad74/attachment.html>
More information about the users
mailing list