For TAs

What to Tell Your Students

Students need to know that the cluster resources should only be used for the course. If students run out of GPU time for a course they need to contact a TA. TAs can ask our service desk to increase time for given students.

Before the new semester starts, access to the cluster is disabled and all home directories are deleted as well as all data under /cluster/courses. Tell your students to copy away their data in time and make sure that data in your course directory is also copied away for use then the course is held again.

Running Jobs

For running jobs your students need to be aware of the following:

• The course tag(s).
• Information an anything that you have prepared under /cluster/courses and how to use it.

Jobs can be canceled by JupyterHub sessions if the cluster is full. These jobs will automatically restart when there are enough nodes free again. This is very unlikely though as the cluster in the past has never been used to 100% capacity.

To avoid running the whole job from the beginning students need to know techniques to use checkpoints or save state in order to resume from somewhere shortly before the job was killed. It is up to you to help them.

JupyterHub

For running Jupyter notebooks on our JupyterHub (if it was requested by your head-TA) students need to know the following:

• The course tag(s).
• Details about the environment(s) under /cluster/courses that you have enabled in the hub.

To use Jupyter with https://student-jupyter.inf.ethz.ch requires you to setup a Jupyter environment under /cluster/courses/{course tag} that we will enable in the chooser for environment of the hub. Instruction can be found here.

Python and Anaconda Environments

We recommend that you set up a python virtual environment or anaconda installation under /cluster/courses/{course tag} with all the packages installed for your particular course. This is especially preferred if you need something like pytorch that requires multiple gigabytes. Then tell your students how to activate the environment.

Students can of course set up their own python environment or install anaconda in their home directory, but that will use up a lot of disk space.

In /cluster/data we keep recent anaconda and miniconda installers ready. Please use these for your class instead of letting each student download them, which would again use unneeded space.

Storage

By default pip and conda cache downloaded packages in the home directory of a user, which quickly fills the home directory. Students only have 20GB of storage.

Please tell your students to use the --no-cache-dir option when using pip. Alternatively pip cache purge can be used to clear the cache after installation.

For conda there is not way to avoid the cache but with conda clean --all the cache can be flushed.

If you provide scripts to your students then please integrate the above to save space in the home directories.

Sharing Data with Students

You can create a folder

which will be automatically linked as {course tag}_data into the home/work directory of any participant that logs in to a login node or starts a Jupyter notebook for your course. Examples, templates and solutions can be placed there for all of your participants. Only participants and TAs of the course can access the data. The data will be read only - participants can copy data such as Jupyter notebooks into their home directory from there to alter them.