Job Submission

Running calculations on Hydra requires submitting a job to the job queue. Hydra uses the Slurm Workload Manager as its job scheduler.

VSCdocRunning Jobs has an extensive introduction to job scripting, submission and monitoring, as well as links to more advance topics. The information in VSC docs covers the classic PBS environment (i.e. qsub, qstat), which is not native to Slurm. However, our cluster at VUB has a compatibility layer that allows using those same commands and job scripts with Slurm. The compatibility is not guaranteed with all possible combinations of features, but most common commands and job scripts should work.

The sections below cover the native tools in Slurm (i.e. srun, sbatch, squeue). All users are encouraged to adopt these tools and use the Slurm commands and options in their job scripts. We provide quick start guides for both new and experienced users who want to migrate to Slurm.

• 1. Slurm Job Scheduler
  • 1.1. Command line tools
  • 1.2. Torque/Moab to Slurm migration
  • 1.3. CPU cores allocation
  • 1.4. Memory allocation
  • 1.5. Slurm partitions
  • 1.6. Job working directory
• 2. Main Job Types
  • 2.1. Serial jobs
  • 2.2. Jobs for GPUs
  • 2.3. Parallel non-MPI jobs
  • 2.4. Parallel MPI jobs
  • 2.5. Hybrid MPI-threading jobs
  • 2.6. Task farming
  • 2.7. Job arrays
  • 2.8. Interactive jobs
  • 2.9. Data management
• 3. GPU Job Types
  • 3.1. GPU generation
  • 3.2. Memory settings of GPU jobs
  • 3.3. Single GPU jobs
  • 3.4. Multi GPU jobs
  • 3.5. Advanced: task distribution in GPUs
• 4. Job Checkpoints and Restarts
  • 4.1. Alternatives to checkpoints
  • 4.2. Checkpoints with DMTCP