The mpi-test_communication test requires excessive resources

[yurivict]

Description

On a computer with 24GB of memory this test causes heavy swapping.

Versions:

• Python version: 3.9
• mpich-3.4.3
• HOOMD-blue version: 3.4.0
• FreeBSD 13.1

[joaander]

Thanks for reporting this behavior. I am not able to reproduce it:

$ /usr/bin/time -v mpirun -n 8 test_communication
        Command being timed: "mpirun -n 8 test_communication"
        User time (seconds): 148.17
        System time (seconds): 3.51
        Percent of CPU this job got: 727%
        Elapsed (wall clock) time (h:mm:ss or m:ss): 0:20.83
        Average shared text size (kbytes): 0
        Average unshared data size (kbytes): 0
        Average stack size (kbytes): 0
        Average total size (kbytes): 0
        Maximum resident set size (kbytes): 279088
...

279088 kiB * 8 = 2.12 GiB. Actual usage will be less than 2.12 GB as ranks 1-7 do not use as much memory as rank 0.

[yurivict]

Apparently all of them run in 2+GB.

[joaander]

Apparently all of them run in 2+GB.

No, not on my test system (Ubuntu 22.04, CUDA 11.7, gcc 11.2). free -m reports 979 MiB more memory available after the test completes compared to when it was running.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.

[yurivict]

not stale

[joaander]

I was unable to reproduce the issue as described. Regardless, HOOMD-blue is a simulation software and should be expected to use large amounts of memory. The unit tests cover a wide range of parameters users may choose which may lead to high memory utilization.