Dear Vasp-users,
We have properly installed the parallel version of Vasp and it was running successfully for the past few months. We had then achieved a fast convergence for our systems containing 198 ions (1038 electrons) and for Ecut=500 eV. However suddenly VASP has slowed down. It shows running status in the queque system but even on 12 processors the OUTCAR file got update only upto 1 ionic iteration after 24 hours.
Is it possible that heavy I/O is going on the cluster machine and thats why this particular program is affected?
We have already cleaned up the share memory. But even after that the program is running slowly. Kindly let us know the probable reasons behind the reduced speed and also suggest possible solutions of overcoming this problem.
Thanking you.
Bhakti
parallel vasp running on slow speed
Moderators: Global Moderator, Moderator
-
bhakti
- Newbie
- Posts: 2
- Joined: Tue Jan 08, 2008 9:09 am
- License Nr.: 836
- Location: Pune, India
- Contact:
parallel vasp running on slow speed
Last edited by bhakti on Mon Jul 28, 2008 9:06 am, edited 1 time in total.
-
forsdan
- Sr. Member
- Posts: 339
- Joined: Mon Apr 24, 2006 9:07 am
- License Nr.: 173
- Location: Gothenburg, Sweden
parallel vasp running on slow speed
If I understand you correctly, you say that you have relaxed your system and now when you want to continue with that run each ionic iteration take much longer? Or is it with slower for all types of calculations?
If it's just this particular run that is slower for each ionic run I would say that the symmetry analysis is different for the continue run than for the initial system due to the relaxation. It's possible that the SYMPREC-tag is to accurate so that it doesn't get all symmetries correctly. If the symmetry is lower the number of irreducible k-points will be higher and thus each ionic step will take longer time. Try to change SYMPREC to see if it will detect more symmtries. Otherwise you can enter the original k-point mesh in that case in the KPOINTS-file if you want to get the same performance as before. However, it depends on the accuracy you want to achieve. For higher accuracy I would keep the new mesh.
However, if you referred to that all runs have beome slower, then this is not likely the reason and more probable some change in your calculation environment. In that case I would ask your administrator if they have made any significant changes to switches, I/O or similar.
Best regards,
/Dan Fors
<span class='smallblacktext'>[ Edited Wed Jul 30 2008, 08:30AM ]</span>
If it's just this particular run that is slower for each ionic run I would say that the symmetry analysis is different for the continue run than for the initial system due to the relaxation. It's possible that the SYMPREC-tag is to accurate so that it doesn't get all symmetries correctly. If the symmetry is lower the number of irreducible k-points will be higher and thus each ionic step will take longer time. Try to change SYMPREC to see if it will detect more symmtries. Otherwise you can enter the original k-point mesh in that case in the KPOINTS-file if you want to get the same performance as before. However, it depends on the accuracy you want to achieve. For higher accuracy I would keep the new mesh.
However, if you referred to that all runs have beome slower, then this is not likely the reason and more probable some change in your calculation environment. In that case I would ask your administrator if they have made any significant changes to switches, I/O or similar.
Best regards,
/Dan Fors
<span class='smallblacktext'>[ Edited Wed Jul 30 2008, 08:30AM ]</span>
Last edited by forsdan on Mon Jul 28, 2008 10:18 am, edited 1 time in total.
-
admin
- Administrator
- Posts: 2921
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
parallel vasp running on slow speed
please also check all .o and .po files, maybe there is a communication (or some other hardware-related) error in this particuar run.
Last edited by admin on Wed Aug 06, 2008 9:49 am, edited 1 time in total.