Usage Policy & Etiquette

The CoSTAR National Lab HPC cluster is a dedicated national computing facility. Because it is a shared resource utilised by researchers and industry partner, maintaining a fair, stable, and highly performant environment relies on good citizenship from all users.

Official Policy Documents

For full institutional context and detailed governance, please refer to the original source documents:

Core Usage Policies

While the cluster serves a national network, its infrastructure is subject to the University of Surrey’s overarching governance.

  • Governing Policies:

    All use of the cluster is subject to the University of Surrey’s IT security and IT acceptable use policies.

  • Zero Tolerance for Abuse:

    Any malicious abuse or misuse of the computing facilities may lead to the immediate suspension of your access. This explicitly includes the detection of any attempt to deliberately circumvent job schedulers to consume more than your entitled share of resources.

  • Emergency Intervention:

    If your jobs are causing significant issues or detrimental impacts to the cluster, they may be stopped by IT services to preserve the system. This action is not punitive; when the system is in danger of crashing, administrators may not have time to notify you before acting.

Note

If you encounter any technical issues or unexpected behavior, please report it promptly via the CoSTAR support portal or Teams channels.

Cluster Etiquette: Being a Good Citizen

Please adhere to the following essential rules for fair and responsible cluster use:

  1. Test code before scaling up: Start small to debug and benchmark before submitting large jobs. Running broken code at scale wastes both your time and valuable cluster resources.

  2. Never manipulate the scheduler: Do not exploit the scheduler to bypass queue limits or priorities. This will get you banned.

  3. Respect shared resources: Avoid hogging CPUs and GPUs, release nodes when your work is done, and clean up your temporary files on shared storage areas.

  4. Avoid wasteful resource requests: Know your workload and specify realistic CPU, memory, and GPU needs in your submission scripts. Requesting massive allocations “just to be safe” clogs the queue for everyone else.

Common Mistakes to Avoid

New users frequently encounter issues by treating the HPC cluster like a standard desktop computer. Avoid these common pitfalls:

  1. Do not run code on the login node directly:

    Intensive tasks must not be run on the head or login nodes. You must use the Slurm scheduler to submit jobs to the compute nodes.

  2. Do not connect VS Code directly to the login server:

    Do not use the standard VS Code Remote-SSH extension to connect to the cluster’s login server. For guidance on this see the VS Code page.

  3. Assume your jobs can be interrupted (Use Checkpointing):

    Cluster operations may be disrupted for various reasons, including technical issues or required maintenance. Always use checkpointing during lengthy simulations or AI model training to save your progress and avoid data loss.

  4. Monitor your storage limits:

    Your home directory has a strict 233GB quota. Always use the high-performance scratch space for writing temporary data during active jobs.

  5. No data on the CoSTAR cluster is backed up!

    You are solely responsible for ensuring that important research data, code, and results are safely transferred off the cluster to your institutional storage or version control systems.