Fair usage

System administration principles

The CLAAUDIA team is responsible for system administration and support. We work closely with the infrastructure team in ITS on maintainance of the system.

The overarching principle in our ressource administration is that we aim to strike a balance between high ressource utilisation, while still leaving the majority of users with the feeling that they can get their work done on the platform.

In time periods of high resource consumption CLAAUDIA may reach out to you and ask about your consumption.

Violations

If we find indications of violations of these principles, CLAAUDIA will contact you to learn more about your situation.

As a general rule CLAAUDIA does not cancel jobs without the user's permission. We do however reserve the right to do so, if the situation is characterised by a degree of severity or urgency, and we can not get a hold of the user.

Ressource consumption

The overall demand for GPU ressources fluctuates throughout the year, and we understand that demand is also dependent on the schedule of the individual researcher. We therefore prefer not to set a fixed limit on how many ressources an individual researcher can consume, as there may be times where they have a legitimate reason for high consumption. Instead we ask our users to be mindful of the overall supply, by making frequent asssesments of the queue.

Responsible ressource consumption also involves taking care with multi-GPU allocations. We encourage our users to experiment with this, but they should take great care to test their applications in order to verify that they can indeed make use of the ressources allocated to their jobs.

If possible, we recomend making good use of times with low consumption - ie. times outside of office hours; on weekends, during holidays, during the night, etc. The parameter --begin can be added to your Slurm command for this purpose.

Not allowed

AI Cloud is designed for processing GPU-demanding batch jobs, that can be executed without the need for user interaction:

Interactive development sessions

By interactive development we mean opening jobs, where you have a GPU available to you, but you only ocassinally run commands on the GPU. This results in a very ineffective utilisation of the GPU's, risks and decreases overall availability. Examples of interactive development session, could be connecting a Jupyter Notebook, Spyder, VS Code to a compute node.

Launching interactive jobs directly in the console is not allowed either - this could be something like: srun --pty -G 1 singularity --nv shell. Going in to an interactive session on a compute node, before launching your job will not release the resources automatically, when they are finished.

Interactive development is allowed on UCloud.

High memory or CPU-demanding operations

If your application can not make use of the GPU's - or does not require one, AI Cloud is not the correct platform for your job. Memory or CPU-demanding operations are currently not allowed on the platform, and we refer to UCloud or LUMI for these purposes. Have a look around this website, or contact CLAAUDIA for guidance on the alternatives.