Hi Jörg, If you don't like to use the MPI backend you can make an easyconfig for GAMESS-US using something else. In that case the toolchain can be lowered as well to a non-mpi one.
The easyblock already supports that, you just need to set the `ddi_comm` option in your easyconfig to sockets. Best, Alex Domingo Scientific Data & Compute VUB - Informatie & Communicatietechnologie Pleinlaan 9 - 1050 Brussel https://hpc.vub.be ________________________________________ From: [email protected] <[email protected]> on behalf of Jörg Saßmannshausen <[email protected]> Sent: Tuesday, 19 December 2023 18:26 To: [email protected] Subject: Re: [easybuild] Reason for no recent US-GAMESS? Dear all, sorry for my lack of engagement in this: snowed under with work. I am not convinced that MPI is good for GAMESS. The last time I checked it used for ever *core* another process which was running only the mpi-dataserver process or whatever it was called. Thus, if you are requesting 24 cores, you are using 12 for GAMESS and another 12 for controlling MPI. For that reason I prefer sockets. https://github.com/easybuilders/easybuild-easyconfigs/pull/16291 The situation might have changed but did anybody actually check? All the best from a mild London Jörg Am 19.12.23 um 15:55 schrieb Alex DOMINGO TORO: > Hi Loris, > > What job scheduler are you using? > > Alex Domingo > > Scientific Data & Compute > VUB - Informatie & Communicatietechnologie > Pleinlaan 9 - 1050 Brussel > https://hpc.vub.be > > > > > ________________________________________ > From: [email protected] <[email protected]> on > behalf of Loris Bennett <[email protected]> > Sent: Tuesday, 19 December 2023 16:44 > To: [email protected] > Subject: Re: [easybuild] Reason for no recent US-GAMESS? > > Hi Alex, > > Alex DOMINGO TORO <[email protected]> writes: > >> Hi Loris, >> >> That should be fixed by my patch >> "GAMESS-US-20230630_add_slurm_support_mpi_target.patch". Are you using >> it in your easyconfig? > > It looks like the patch is being applied: > > ... > == 2023-12-19 16:12:46,736 easyblock.py:2583 INFO Applying patch > GAMESS-US-20230630_add_slurm_support_mpi_target.patch > == 2023-12-19 16:12:46,739 run.py:246 INFO running cmd: patch -p0 -i > /tmp/eb-micogr01/files_pr19310/g/GAMESS-US/GAMESS-US-20230630_add_slurm_support_mpi_target.patch > == 2023-12-19 16:12:46,761 build_log.py:267 INFO preparing... > ... > > but I still get > > ... > setenv GMCCCS /tmp/eb-micogr01/tmpqeov8jwv/exam27.F99 > unset echo > Alas, MPI only supports SGE, PBS, or user-given node lists. > Please locate this place in the script and edit it. > (at > easybuild/software/EasyBuild/4.8.2/lib/python3.6/site-packages/easybuild/tools/run.py:681 > in parse_cmd_output) > == 2023-12-19 16:33:14,534 filetools.py:2012 INFO Removing lock > /trinity/shared/easybuild/software/.locks/_trinity_shared_easybuild_software_GAMESS-US_20230630-R2-gompi-2022a.lock... > > Any ideas? > > Cheers, > > Loris > >> >> Alex Domingo >> >> Scientific Data & Compute >> VUB - Informatie & Communicatietechnologie >> Pleinlaan 9 - 1050 Brussel >> https://hpc.vub.be >> >> >> >> >> ________________________________________ >> From: [email protected] >> <[email protected]> on behalf of Loris Bennett >> <[email protected]> >> Sent: Tuesday, 19 December 2023 14:34 >> To: [email protected] >> Subject: Re: [easybuild] Reason for no recent US-GAMESS? >> >> Hi Alex, >> >> Alex DOMINGO TORO <[email protected]> writes: >> >>> Hi Loris, >>> >>> I got the same errors with libxc v6.2.2 in gompi/2023a. GAMESS-US >>> v2023 does not yet support the newer libxc v6. So I pushed this >>> easyconfig of gamess down to `gompi/2022a` where we still have libxc >>> v5. >>> >>> libxc is indeed optional though, so it is also possible to make an >>> easyconfig without libxc in `gompi/2023`. But my users want it, so I >>> have to add it. >>> >>> I also fixed the issues with the tests and now the builds with EasyBuild >>> are working. >>> >>> You can find my easyconfigs for GAMESS-US v20220930 and v20230630 in >>> https://github.com/easybuilders/easybuild-easyconfigs/pull/19310 >> >> Thanks for the new EC. The build seems to go OK, but I get the >> following in the test step: >> >> ... >> setenv GMCCCS /tmp/eb-ztp6ke_3/tmpzifeb6n8/exam27.F99 >> unset echo >> Alas, MPI only supports SGE, PBS, or user-given node lists. >> Please locate this place in the script and edit it. >> (at >> easybuild/software/EasyBuild/4.8.2/lib/python3.6/site-packages/easybuild/tools/run.py:681 >> in parse_cmd_output) >> == 2023-12-19 14:19:01,748 filetools.py:2012 INFO Removing lock >> /trinity/shared/easybuild/software/.locks/_trinity_shared_easybuild_software_GAMESS-US_20230630-R2-gompi-2022a.lock... >> == 2023-12-19 14:19:01,748 filetools.py:383 INFO Path >> /trinity/shared/easybuild/software/.locks/_trinity_shared_easybuild_software_GAMESS-US_20230630-R2-gompi-2022a.lock >> successfully removed. >> == 2023-12-19 14:19:01,748 filetools.py:2016 INFO Lock removed: >> /trinity/shared/easybuild/software/.locks/_trinity_shared_easybuild_software_GAMESS-US_20230630-R2-gompi-2022a.lock >> == 2023-12-19 14:19:01,748 easyblock.py:4277 WARNING build failed (first >> 300 chars): cmd "export OMP_NUM_THREADS=1; SCR=/tmp/eb-ztp6ke_3/tmpzifeb6n8 >> /trinity/shared/easybuild/software/GAMESS-US/20230630-R2-go >> mpi-2022a/rungms exam27 20230630-R2 8 8" exited with exit code 9 and >> output: >> 1 : exam27 >> 2 : 20230630-R2 >> 3 : 8 >> 4 : 8 >> ... >> >> I am not sure what the line >> >> Alas, MPI only supports SGE, PBS, or user-given node lists. >> >> is trying to tell me. >> >> [snip (182 lines)] >> >> Cheers, >> >> Loris >> >> -- >> Dr. Loris Bennett (Herr/Mr) >> FUB-IT (ex-ZEDAT), Freie Universität Berlin > -- > Dr. Loris Bennett (Herr/Mr) > FUB-IT (ex-ZEDAT), Freie Universität Berlin >
