HPC software & scheduler setup

We map departments, projects and workloads to an operating policy before building queues. The discussion covers permitted job duration, resource limits, priority, reservations and what happens when demand exceeds capacity. A small teaching exercise and a long simulation may need different access rules even when they share the same physical cluster.

Slurm is an option for managing Linux cluster resources and queued batch jobs. The scope defines controller services, compute-node registration and the resource model that will be exposed to users. Configuration is documented alongside operating-system dependencies, authentication requirements and the procedure for taking a node out of service.

Researchers need compatible compilers, libraries and application versions. We agree how environments will be selected and how changes will be introduced without silently breaking existing jobs. Licensed scientific applications require their own entitlement and licence-server arrangements. Example submission scripts are prepared for the applications included in the project scope.

A browser portal can be assessed where researchers need interactive access without learning every command-line workflow. Acceptance checks include a submitted job, queue limits, output retrieval and controlled job cancellation. The operating handover explains how administrators investigate failed work, manage users and communicate planned maintenance.

Pricing depends on node count, user groups, applications, identity integration and the access tools required. Application porting, extensive performance tuning and commercial licences are separately scoped. Share your cluster inventory and sample job requirements so the quotation can distinguish a basic scheduler rollout from a wider research computing platform.