[slurm-users] squeue: node vs. core memory request

[Oleksandr Moskalenko]

Hi,

I wonder if anyone was able to address the following squeue issue - 
whether with a workaround or with a patch.

The squeue command's %m format type specification is described as 
providing a 'Minimum size of memory (in MB) requested by the job.' 
Unfortunately, as far as I can tell it does not appear to be possible to 
differentiate between mem-per-cpu and node memory requests from the 
returned information making it impossible to determine the sum of all 
memory requests in pending and running jobs for a user or an account 
from squeue output.

On the contrary, the format used by sacct does differentiate between 
those two types of requested memory as

ReqMem - "Minimum required memory for the job, in MB. A 'c' at the end 
of the returned number represents Memory Per CPU and 'n' represents 
Memory Per Node. Note: This value is only from the job allocation, not 
the step."

Fixing this deficiency in squeue would be ideal as it would allow 
accurate calculation of memory requests in running and pending jobs, 
which is critical for throughput optimization on SLURM clusters, but a 
workaround would be great in short-term as well.

Cheers,

Alex