You are here: Home / News & Events / News / Responsible use of mistralpp.dkrz.de

Responsible use of mistralpp.dkrz.de

Dec 15, 2020

Dear users,

Some of you using the interactive mistralpp nodes may have noticed that we often have to restart those nodes because of resource overload.

On mistralpp nodes, we do not enforce the restrictions we put on the login nodes for mistral. This allows you to run more demanding tasks. While we allow this type of usage, you should keep in mind that other users are working on the node as well and resources are limited. If the node crashes because you overloaded its memory or CPUs, all your colleagues on that node risk loosing their current work.

We therefore urge you to show good judgement when using mistralpp. The nodes have 24 CPU cores and 256 GiB of memory. While this may seem much compared to your local laptop or desktop computer, you have to share the node with about 50 other users.

For many tasks, an allocation or job is the better alternative. You can find example job scripts on our website and the login banner on mistralpp has examples for using dedicated resources with salloc. Using jupyterhub.dkrz.de may also be a good alternative since kernels for JupyterHub run in a job on dedicated resources. Apart from Python we now also provide kernels for Julia and R.

The successor to mistral, levante, will only offer compute resources through the Slurm workload manager. So now may be a good time to start restructuring your workflows if they still depend on mistralpp.

Best regards,
DKRZ

Document Actions

Filed under: