Jobs on Atlas

Atlas uses the Slurm Workload Manager as a scheduler and resource manager.

In order to run a job, users should do the following from the log-in node

• Request resources
  • Parameters
    • -N - number of nodes needed
    • --mem - amount of RAM needed per node
    • -n - number of cores needed per node
    • --time - amount of time needed per node
  • Commands
    • salloc - interactive use
    • sbatch - queued job
      • Parameters can be included in script
• Provide commands/scripts to be executed on allocated resources
  • srun - interactive use
  • sbatch - queued job

Job Arrays

Job arrays are a mechanism for managing a set of jobs that have identical resource requirements.

• Created with
  • --array 0-2 (creates 0, 1, 2)
  • --array 0, 2, 4 (creates 0, 2, 4
  • --array 0-6:2 (creates 0, 2, 4, 6)
• Each job in the array knows which part of the array it is
  • SLURM_ARRAY_TASK_ID
• Example: Parsing 12 datasets
  • Create job array with --array 0-11 when submitting script
  • Script contains a list of data
  • Each script contains a command to parse the SLURM_ARRAY_TASK_IDth dataset