Child pages
  • Summary of the tutorial
Skip to end of metadata
Go to start of metadata
  • You can compile, edit scripts and view results on the login nodes, but computational work should be run on the compute nodes
  • You can access compute nodes with srun
    • Either via a job script, or interactively
    • Compute nodes are allocated to jobs by the scheduler, so your job might not start immediately
    • Jobs must request resources, but mostly need not specify a queue.
    • Requesting just slightly more than when you expect to need is generally the best practice
    • Short jobs get higher priority, and short or small jobs are easier to schedule quickly
  • You can monitor your job's progress with squeue, sstat, sacct, scontrol or slurmtop
  • Software is managed by Environment Modules
    • Use 'module avail' to find software packages
    • And 'module load' to load them into your environment
      • including within job scripts!
    • Use 'module purge' to return to a clean environment before loading a new set of modules
    • Other useful commands are 'module list' and 'module show'

 

  • No labels