Leonhard

isginf facilitates access to two shares on the Leonhard cluster operated by ITS, one for researchers and one for courses.

Department Share

The department share is is set up to use the idle time of the student share (see below). The department share at the moment does not have its own nodes. It also has no restrictions on the number of GPUs per user.

Users of the department share have no on cluster storage, only the home directory (16Gb). If cluster storage is needed it has to be organized by the research group. isginf can help with the procurement.

Access

Access is granted on demand and only when needed and for members of the department. Individual members of the department can contact our service desk. ITCs can request access for their research group as a whole and will receive a group to manage access themselves.

Adding undergraduate students working on projects to the department share is discouraged. If needed they can be added to the student share.

If we find that long-running jobs of researchers block our students from doing their exercises we may have to restrict free access again.

Usage

Once a user has access, running jobs works as described in the scientific computing wiki.

User who have access to more than one share, for instance users of research groups with their own Leonhard share, can run jobs in the department share with the following command:

bsub -G ls_infk ...

Student Share

The student share currently consists of nine nodes with a total of 72 GPUs and 20TB fast storage. Usage of these nodes is restricted to courses that require GPU computing. The Euler cluster should be used for normal CPU based jobs, every student has access there.

Access

Access is managed through a group for each course, only members of a course group can log in. The group can be managed by the head TA of the course as well as the ITC. With a list of student ETH email addresses isginf can initially set up a group at semester start.

Rules

For students the usage rules are as follows:

  • Only one concurrent job on one GPU is possible per user.
  • If you are grouped in teams, only one member of the team may submit a job at a time. For large classes there are not enough GPUs for each user, only for each team.
  • At the end of a class you will lose access to the cluster and your home directory and will be deleted. Please be sure to copy away all files you want to keep.

For collaboration among team members we offer to provide GitLab for courses.

Documentation

All documentation on the cluster itself is maintained by ITS and can be found on their scientific computing wiki.

Log in via ssh with your ETH user name to login.leonhard.ethz.ch to get started.

Page URL: https://isg.inf.ethz.ch/bin/view/Main/ServicesClusterComputingLeonhard
2019-12-13
© 2019 Eidgenössische Technische Hochschule Zürich