[slurm-users] Slurm and memory
Dmitri Chebotarov
dchebota at gmu.edu
Mon Apr 9 08:39:12 MDT 2018
Hello
I'm trying to figure out how to change SLURM's behavior on gathering free memory from nodes.
At this time 'sinfo' reports 'free' memory from the node (and not 'available'):
F.e:
#sinfo -eN -o %N,%m,%e,%C
...
NODE067,64170,14672,0/32/0/32
I can see the NODE067 has no jobs running - 0 CPUs allocated, but SLURM reports 14672 out of 64170 is available, while in fact 'free -m' command reports '62830' as available.
[dchebota at NODE067 ~]$ free -m
total used free shared buff/cache available
Mem: 64170 704 14669 35 48795 62830
Swap: 10239 0 10239
Is it possible to change SLURM's behavior to read 'available' instead of 'free'?
'buff/cache' is considered as available memory and will be allocated when needed.
So far, I need to run 'sync && echo 3 > /proc/sys/vm/drop_caches' to clear 'buff/cache' as workaround.
Thank you,
--
Dmitri Chebotarov.
George Mason University,
4400 University Drive,
Fairfax, VA, 22030
GPG Public key# 5E19F14D: [https://goo.gl/SlE8tj]
More information about the slurm-users
mailing list