From TheoChem Cluster
Jump to navigation Jump to search

Submitting a job

The queuing system on TheoChem is SLURM and it works similar to peregrine.

  • sbatch to submit a job;
  • scancel to cancel a job;
  • squeue to see what is in the queue (use man sbatch to see the manuals).

We have the following queues (partitions) on the TheoChem Cluster:

  • ultrashort: special queue for ultra short jobs (< 30 minutes);
  • short: default queue for short jobs (< 1 days);
  • medium: special queue for very medium jobs (< 3 days);
  • long: special queue for long jobs (< 10 days);

All queues have:

  • default wall time 00:05:00
  • default mem_per_cpu 2048

Job example

A typical job looks like:

#SBATCH --time=0:30:00
#SBATCH --partition=short
#SBATCH --mem-per-cpu=3GB
#SBATCH --ntasks-per-node=28
#SBATCH -o out.o%j

mpirun -n 56 ./a.out 

  • A wall time of 30 minutes is chosen in this example;
  • The job is supposed to run on 2 nodes;
  • The job uses on each node 28 cores;
  • The short partition is chosen;
  • The output will be called out.o(job_id).