Parallelism problem in Response calculation in Y-4.3.2

Deals with issues related to computation of optical spectra, in RPA (-o c) or by solving the Bethe-Salpeter equation (-o b). Includes local field effects, excitons, etc.

Moderators: Davide Sangalli, andrea marini, Daniele Varsano, andrea.ferretti, myrta gruning, Conor Hogan

Post Reply
sitangshu
Posts: 119
Joined: Thu Jan 05, 2017 8:08 am

Parallelism problem in Response calculation in Y-4.3.2

Post by sitangshu » Wed May 29, 2019 1:56 pm

Dear Sir,

I am experiencing a parallelism problem with the calculation of static dielectric function with the version yambo 4.3.2
Strangely, after the last response calculation:
Xo@q[144] |####################################### | [097%] 54m-04s(E) 55m-26s(X)
the code stops doing nothing, unless it is killed.
I am using a parallel strategy as 128 1 2 1 1 and compiled the code as:
./configure --enable-mpi MPIFC=mpif90 MPIF77=mpif77 MPICC=mpicc --enable-3d-fft --with-p2y-version=qexsd --enable-dp

I have checked with other CPU combinations, but same problem appears. A single node calculation with a distribution 32 1 1 1 1 runs smoothly for lower parameter thresholds.

I have seen a similar problem reported here:
viewtopic.php?f=13&t=1563
But the solution is not mentioned properly. Can you help me on this? The problem did not appear for lesser yambo versions.

With regards,
Sitangshu
Sitangshu Bhattacharya
Indian Institute of Information Technology-Allahabad
India
Web-page: http://profile.iiita.ac.in/sitangshu/
Institute: http://www.iiita.ac.in/

User avatar
Daniele Varsano
Posts: 2712
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Parallelism problem in Response calculation in Y-4.3.2

Post by Daniele Varsano » Wed May 29, 2019 2:12 pm

Dear Sitangshu,
not easy to understand the problem if any error is emitted.
The only suggestion I can provide, in case the problem is due to a memory issue is to set:
PAR_def_mode= "memory"

or alternatively choose a strategy assignign the cpu's to the bands (c,v). This will optimize the memory per core needed.
Other suggestion is to compile the code with the option:
--enable-memory-profile

this will give information on the memory needed per proc in the log file.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

Post Reply