[gmx-users] Known but still domain decomposition

Alex alexanderwien2k at gmail.com
Wed May 2 11:33:25 CEST 2018


Deal all,
I am doing coarse-grained bonded and nonbonded parametrization using
votca-gromacs. The "mpirun -np 64 gmx_mpi mdrun" work fine till step_021 of
parametrization where it crashes because it can not find any domain
decomposition for 64 ranks that is compatible with the system size. Even it
works in the minimization simulation in step_021.
I have been reading most of the previous discussions about the error of
domain decomposition, and I have tried different options like using
different ranks (64, 32, 16, 8, 40, 36) and also -rdd ... or -dds. However,
no success yet.
About the machine: Running on 4 nodes with total 64 cores, 64 logical
cores, Cores per node: 16.

Below I have shared the md.log files for the step_20 (which works fine) and
step_21 (minimization which works fine) and step_21 (which crashes), and I
would be so appreciated if one could help me overcome the issue.

md.log-20
https://drive.google.com/open?id=1ZwYggDI1Hy_eYY8_nT8I50t91Wla0Obr

md.log-21-mini
https://drive.google.com/open?id=1oYE4A6fC1L8u_vx4sR8RIq3EWiG9-7YB

md.log-21
https://drive.google.com/open?id=1ff_QJrdGvLFgwGU9xMY-IcqfxzF2jkAO

Thank you.
Regards,
Alex


More information about the gromacs.org_gmx-users mailing list