Job Submission

Running calculations on the VUB-HPC clusters requires submitting jobs to the Slurm Workload Manager.

Slurm schedules these jobs and distributes them across the cluster, ensuring that resource usage is maximized while maintaining fair usage among all users.

A comprehensive introduction to job scripting, submission, and monitoring, along with links to more advanced topics, can be found in the VSCdocRunning Jobs section.

The sections below cover specific job submission use cases for the VUB-HPC clusters, ranging from parallel execution and GPU usage to detailed instructions for specific scientific software available in the cluster.

• 1. Slurm Workload Manager
  • 1.1. Command line tools
  • 1.2. Job working directory
  • 1.3. Torque/Moab to Slurm migration
• 2. Slurm clusters
  • 2.1. Hydra: main cluster
  • 2.2. Anansi: test/debug cluster
• 3. Slurm partitions
• 4. Resource allocation
  • 4.1. CPU-cores allocation
  • 4.2. Memory allocation
• 5. Main Job Types
  • 5.1. Serial jobs
  • 5.2. Jobs for GPUs
  • 5.3. Parallel non-MPI jobs
  • 5.4. Parallel MPI jobs
  • 5.5. Hybrid MPI-threading jobs
  • 5.6. Task farming
  • 5.7. Job arrays
  • 5.8. Interactive jobs
  • 5.9. Data management
• 6. GPU Job Types
  • 6.1. GPU generation
  • 6.2. Memory settings of GPU jobs
  • 6.3. Single GPU jobs
  • 6.4. Multi GPU jobs
  • 6.5. Advanced: task distribution in GPUs
  • 6.6. Shared GPUs on Anansi
• 7. Job Checkpoints and Restarts
  • 7.1. Alternatives to checkpoints
  • 7.2. Checkpoints with DMTCP