Dear Prof. Blaha, Thank you for the explanation. The general idea was hard to get from the UG. Sincerely Mikhail >Пятница, 15 октября 2021, 19:37 +03:00 от Peter Blaha ><pbl...@theochem.tuwien.ac.at>: > >I think (hope) the parallelization of this script is well described in >the UG or simply in the "online help" using -h switch. It has multiple >options and levels for parallelization: > >psi11:/psi11/scratch> optimize_abc -h >USAGE: optimize_abc [-h -t 2/3 -sp -p -n X -FC X -d X -ctest X Y Z >-ana X -j "run_lapw -p ..." ] >optimizes a,(b),c lattice parameters >-p requires the presence of .machines (single jobstep) and > .machines_1...4 (9) for 4 (9) parallel jobsteps in the 2D (3D) case > >The script makes a scf calculation for the present lattice parameter in >the case directory. This calculation uses the standerd .machines file >when specifying "run_lapw -p" as job. > >However, then it has to make changes in 4 (or 9 for the 3Dcase) >directions. This can be done in serial or in parallel (using the -p >switch of optimize_abc). So with -p it will span 4 (9) run_lapw jobs in >parallel. >If you still have more cores available, you can in addition supply >.machines_1, .machines_2, ...4 (9) files. > >So suppose you have 4 nodes with 16 cores each, you could put into each >of these .machine_X files 16 different cores (eg. in mpi), but run 4 mpi >jobs in parallel. >In addition you create a .machines with all 64 cores for the "starting >job" (at least if it is still efficient for your example. Remember: a >very small cell will run MUCH LONGER in mpi with 64 cores (or even >crash) then on fewer cores. > >The "task" parallelization is MUCH more efficient then heavy mpi >parallelization. > > > >Am 15.10.2021 um 17:28 schrieb Mikhail Nestoklon: >> Dear wien2k community, >> I am trying to use new script optimize_abc_lapw on a cluster. Something >> in its behavior in terms of computer power consumption confused me and I >> am checking how it actually works. I realized that at some point (at >> least when ‘doing x-zchange’) it runs lapw0 and lapw1c and not >> lapw0_mpi, etc. The most strange part is that when it starts it >> correctly uses mpi versions of the programs. >> Is this correct behavior? >> I run the script as ‘optimize_abc_lapw -p’ at the end of slurm script >> which prepares .machines file. >> The structure is hexagonal. >> >> Thank you in advance. >> Sincerely yours, >> Mikhail Nestoklon >> >> _______________________________________________ >> Wien mailing list >> Wien@zeus.theochem.tuwien.ac.at >> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien >> SEARCH the MAILING-LIST at: >> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html >> > >-- >-------------------------------------------------------------------------- >Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna >Phone: +43-1-58801-165300 FAX: +43-1-58801-165982 >Email: bl...@theochem.tuwien.ac.at WIEN2k: http://www.wien2k.at >WWW: http://www.imc.tuwien.ac.at >------------------------------------------------------------------------- >_______________________________________________ >Wien mailing list >Wien@zeus.theochem.tuwien.ac.at >http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien >SEARCH the MAILING-LIST at: >http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
_______________________________________________ Wien mailing list Wien@zeus.theochem.tuwien.ac.at http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien SEARCH the MAILING-LIST at: http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html