Hello,
I am not a very experienced user of ABINIT code and I tried, following the e-ph tutorial, to compute some supercell calculations (w/ and w/o vancacy), so I need to be able to use parallelisation. I found that I could include parallelisation on the perturbation levels and I tried to couple it with bands and k point parallelisation, it seems to be working until I notice the warning in the my log files. The warning screenshot follows attached.
I would gladly receive any indications on the best way to procede to use parallelisation in my situation. I tried a general band, fft and kpoint general parallelisation, after ensuring that the band and k point would be the same in all the datasets along with the perturbation parallelisation in the DS where I calculated those. Though I am not sure, after noticing this warning.
Thank you. Best regards,
Rafaela
Parallelisation on multidataset RF
Moderators: mverstra, joaocarloscabreu
Parallelisation on multidataset RF
- Attachments
-
- Screenshot 2017-08-21 15.03.55.png (16.48 KiB) Viewed 2333 times
Re: Parallelisation on multidataset RF
Things happen automatically, you don't need to add the input variables. In particular kpt and bands are distributed. You can check explicitly with an example: run the same quick input with 2,4,8 processors and you should see a speedup (unless it is really too small a job and the i/o and initialization is dominant).
If you have a recent version of abinit (please read the nettiquette and give us some additional information) the fft is also beginning to be parallelized, but that depends on your system size and version of the code.
If you have a recent version of abinit (please read the nettiquette and give us some additional information) the fft is also beginning to be parallelized, but that depends on your system size and version of the code.
Matthieu Verstraete
University of Liege, Belgium
University of Liege, Belgium