Problem with paral_kgb

option, parallelism,...

Moderators: fgoudreault, mcote

Forum rules
Please have a look at ~abinit/doc/config/build-config.ac in the source package for detailed and up-to-date information about the configuration of Abinit 8 builds.
For a video explanation on how to build Abinit 7.x for Linux, please go to: http://www.youtube.com/watch?v=DppLQ-KQA68.
IMPORTANT: when an answer solves your problem, please check the little green V-like button on its upper-right corner to accept it.
Locked
jjlow@yahoo.com
Posts: 1
Joined: Fri Mar 19, 2010 7:18 pm

Problem with paral_kgb

Post by jjlow@yahoo.com » Fri Mar 26, 2010 6:40 pm

I have configured and built a parallel abinit executable with the Intel compilers (version 11.1.059), the mkl libraries (10.2.2.025) and mvapich2.

I have successfully run the serial tests provided with the code. I have run several successfull tests of parallelism over kpoints and planewaves.

However, I have not been to run parallel over kpoints, planewaves and fft.

I would like to use abinit on a computer with over two thousand cores and fast interconnects. Parallelization over kpoints, planewaves and fft would help alot.

Could anyone help me understand why my abinit executable will not run with paral_kgb=1 (parallel over kpoints, planewaves and ftt)?

I have attached the log file from the failed si_kpt_band test case and the script I used to build the parallel version of abinit.

Thanks

John J. Low
Argonne National Laboratory
Attachments
log_si_kpt_band_fft.out
log from si_kpt_band test case
(89.8 KiB) Downloaded 289 times
config-netcdf.out
script to configure and make parallel version of abinit
(704 Bytes) Downloaded 275 times

fbottin
Posts: 1
Joined: Tue Dec 01, 2009 9:24 am

Re: Problem with paral_kgb

Post by fbottin » Thu Apr 08, 2010 4:50 pm

Dear John,

I don't really understand your problem.
The log file enclosed in your post contains 8 datasets and a few of them involve 1 k-point only.
So you can't parallelize over k-point using this kind of input file.
In order to test the triple kpt-band-fft parallelisation I suggest to consider the 8th dataset (with 30 k-points), and only this one.
Set for instance npkpt=10, npband=6 and npfft=3 (these dimensions dividide nkpt, nband and ngfft, respectively).
For more details about these input variables, see http://www.abinit.org/documentation/hel ... #paral_kgb,
and for more details about 3-levels calculations, see our paper:
F. Bottin, S. Leroux, A. Knyazev and G. Zerah, Large scale ab initio calculations based on three levels of parallelization, Comput. Mat. Science 42, 329 (2008) (available on arxiv: http://arxiv.org/abs/0707.3405).
For this 3-levels calculation you need 180 procs at all, which is a good test for a first 3-levels calculation.
And, you sould obtain a very poor scaling since the system is very small ;)

Regards,
Francois Bottin

NB: I suggest also to switch to wfoptalg 14 (more efficient and more stable).

Locked