[Nebula-users] Memory available to jobs decreased

[Peter Bortas]

Dear Nebula users,

The amount of memory available to Slurm jobs has been decreased by
2GiB per node, aka 32KiB per thread.

For most (probably all) of you this probably doesn't matter, but jobs
that where extremely close to the memory limit already might start
failing due to out of memory.

Background: This has been done to leave some more space for the
operating system to work. Some recent filesystem OS updates requires
more RAM to remain stable, and we've had several incidents where jobs
running out of RAM has also locked up the node so it has to be
rebooted.

God Jul,
--
Peter Bortas, NSC