[lammps-users] Segmentation Fault Error

Dear colleagues,

I was testing a simulation on LAMMPS 29Oct20 in a new cluster and on this new occasion I got a segmentation fault error.

I would like to know if this error is related to the mpi version installed on the cluster (I suppose that could be it, from previous forum responses), a memory error or if my simulation is “quite inefficient” (128 MPI tasks x 1 OpenMP threads for a system of 2837 atoms).

Attached, I send the “error.err” file (error file produced by the cluster’s queuing system) and the “output.out” (generated by LAMMPS).

If someone with more experience could kindly take a brief look at these files and tell me if the problem is indeed mpi related, I would be grateful.

Thanks in advance for your time,

Best regards.


error.err (13.9 KB)

output.out (53.7 KB)

using that many MPI ranks for so few atoms is most certainly wasteful.
your input is comparable in terms of the computational requirements to the “rhodo” benchmark input form LAMMPS.
If you look at the “fixed size” benchmark graph here: https://lammps.org/bench.html#rhodo
You see a significant drop in parallel efficiency for more than 64 MPI ranks. and that is for an input with 32000 atoms.
You have 2837 atoms.

Your timing breakdown also tells a story of inefficiency and waste:

MPI task timing breakdown:
Section | min time | avg time | max time |%varavg| %total

Thank you very much Dr. Axel Kohlmeyer, your comments are always surgical!

Best regards.

Have a nice day,