HPC for Science

Building resources to support science isn't a lot different from building resources to support traditional commercial industries that use HPC resources, aside from your customer base. In a traditional commercial environment (oil exploration, finance, modeling and design, etc.), your customer base usually runs a small set of software products, maybe even a single program. In a science/research environment, the software base that you need to support never seems to stop growing.

This section covers some common issues in designing and configuring an HPC resource.

Structure of a basic cluster.

Common Job managers/Batch schedulers

  • Slurm
  • Maui/Torque
  • Moab
  • PBS
  • Sun Grid Engine
  • LSF

Scheduling algorithms

  • FIFO with Backfill
  • Fairshare/Fair Tree

Parallel file systems

  • Lustre
  • PVFS/OrangeFS
  • BeeGee FS
  • Ceph?

Cloud Computing

Working with scientists

Working with System Administrators

Reproducability

Managing User Environments