We use cookies to ensure that we give you the best experience on our website. You can change your cookie settings at any time. Otherwise, we'll assume you're OK to continue.

Durham University


SLURM hints and tips

To use fewer than all cores in a node, with a multiple node job, use something like:

(e.g. for cosma6, requiring 256 MPI tasks (but 512 cores), on 32 nodes, so 8 MPI ranks per node - cosma6 has 16 cores per node)

> #SBATCH --ntasks=256
> #SBATCH --cpus-per-task=2

Batch Queue priorities

The priority of pending jobs in a queue is determined by a number of factors. These include:

1. How successful the user has previously been at getting time (which is a factor that decays over time)

2. Group fair-share: this can be raised for small groups/projects

It is possible to change the priority of you jobs:

scontrol update job JOBNUMBER nice=100 (which will lower its priority)

Queue information

The showq command can be used to obtain useful information about the queues (in addition to squeue and sinfo). For example:

showq -q cosma7 -l -f -o

The c7jobload command will give the load (i.e. CPU demand) of a job. Use -h to see options.

sinfo -s will give a suscinct overview of the partitions.

The sstat (for running jobs) and sacct (for completed jobs) provide information about the jobs. For example, at the end of your batch script, if you use:

sstat --jobs=${SLURM_JOBID}.batch --format=jobid,maxrss,ntasks

it will print information about memory used.