Cluster Etiquette

Please be considerate of other cluster users, in particular please keep the following in mind.

Remember that you are not the only person using this cluster.
You are not the only one with a paper deadline.
If you are unsure what you are doing please ask.
Profile your jobs and spend a little time figuring out how best to run them.
Pay attention to how much memory your job needs.
Please test your jobs first before submitting hundreds or thousands of them.
Please ensure you make good use of GPUs. Check that your computations are really running on the GPU.
Never set a script to submit jobs into the queue in large numbers, use array jobs instead.
If you submit a job that seems to break things please email cluster-support@cs.ucl.ac.uk and let us know, we will try to help.
Please ensure that you are not repeating someone else in your group’s work, just so you don’t have to name them on your paper, talk to each other.
Do not run code directly on compute nodes, qsub or qrsh everything.
Do not be lazy just because we have a lot of processing power. The HPC is a costly resource.
Do not run jobs that require high disk I/O over NFS. Request some scratch space with -l tscratch=xG then copy your data to /scratch0
Request a project store for you project data.
Do not use cluster file stores for storing inappropriate data.
Do not use the processing power for “personal projects”, it is a research cluster.
If this doesn’t make sense to you then arrange to attend a cluster induction.

Anyone caught wilfully abusing the cluster will have their account suspended. Serious malpractice may lead to suspension from UCL. Usage of the cluster is subject to the usual T&C’s for Computer Science computers. If you are writing grant proposals please add in some compute node costs so that the cluster can continually be upgraded. Please see TSG for an idea of costs.

If you would like to contribute to these web pages in any way please contact us.