A few things to add to the documentation after the November users committee

[lesteve]

@remicres @mdiazmel feel free to edit and add more if I missed some:

• ability to ask for CPU hours (it should stay a small fraction of GPU hours) in Dynamic Access. I guess we should try to check that we can find it in the eDARI form and do a screenshot or something
• compil machines. My understanding is that those are interactive machines where some of the limitations of the login node does not apply. They were created mostly to do multi-thread compilation because on the login nodes you are restricted to use only one CPU via cgroups. Hopefully we'll have more details about these machines in one of the next Flash Info
• datasets legal stuff for Health-related datasets. Jean Zay does not have the label "Hébergeur de Données de Santé" so this adds complication for Health-related datasets. You need to contact user support if you are unsure about one of your dataset. This may take something like a month to get approval (dataset-by-dataset basis) with some administrative work with some legal departments involved.

[RemiLacroix-IDRIS]

compil machines. My understanding is that those are interactive machines where some of the limitations of the login node does not apply. They were created mostly to do multi-thread compilation because on the login nodes you are restricted to use only one CPU via cgroups.

That's correct, except that they are not really "interactive" (you cannot SSH on them directly). You need to request some resources via Slurm first. The easiest way is to use something like srun -c 5 -p compil --pty bash.

We have added some info on our documentation (for example in http://www.idris.fr/eng/jean-zay/cpu/jean-zay-cpu-exec_partition_slurm-eng.html) and it will be in the next Flash-Info.

[lesteve]

OK thanks for the details! It is really great that you don't need CPU hours to use them.

On top of this, these machines have access to the internet. This is great too because I have heard of workflows (e.g. PlantNet) where they need to refresh the dataset from their external databases and they need both CPUs and internet access.