Issue using network processing features on HPC cluster with SLURM

[fwautele]

Hi all,

We plan to use Photoscan network processing features on a HPC cluster. The scheduler used is SLURM (https://slurm.schedmd.com). Linux cgroups are used to contraint memory and cpu on compute nodes (ProctrackType=proctrack/cgroup and TaskPlugin=task/cgroup in slurm.conf).

When running Photoscan in "processing node" mode directly on compute node, all the cores are used by batch processes but when submitting through SLURM, only one core is used per task.

Is multithreading disabled when running Photoscan in cgroups?

Thanks!

With best regards