[Beowulf] [OOM killer/scheduler] disabling swap on cluster nodes?

Christopher Samuel samuel at unimelb.edu.au
Tue Feb 17 15:53:34 PST 2015


On 10/02/15 17:14, Mark Hahn wrote:

> it's also worth pointing out that you can explicitly
> tell the OOM killer to lay off a process (/proc/$pid/oom_*).
> sshd does this to itself, for instance - it would make sense for a
> scheduler daemon to do so as well.

The Slurm daemon that lives on the nodes already does this (code
contributed by NUDT in China in 2009).

 *  set_oomadj.c - prevent slurmd/slurmstepd from being killed by the
 *      kernel OOM killer

-- 
 Christopher Samuel        Senior Systems Administrator
 VLSCI - Victorian Life Sciences Computation Initiative
 Email: samuel at unimelb.edu.au Phone: +61 (0)3 903 55545
 http://www.vlsci.org.au/      http://twitter.com/vlsci



More information about the Beowulf mailing list