Skip to content

Using the MaRC3a cluster

Prerequisites

To use the MaRC3 HPC cluster in the DataHub setting, the following prerequisites must be met:

  • Register for the DataHub and acknowledge the stated terms of use and legal regulations. If your registration is positively reviewed, the DataHub Team will provide you with the required Uni Marburg staff account and HPC permission.
  • Connect to the Uni Marburg staff VPN, if you are not directly connected to the Marburg University network (two factor authentication is required).
  • Configure your shell to be bash via the web form: https://admin.staff.uni-marburg.de/aendshell.html
  • Configure Git on the HPC as described in the Setup and Installations section of the this workflow tutorial. Git and Git-LFS are already installed on the cluster and are enabled by default.
  • We recommend you to attend the regular MaRC3 cluster introduction and/or to work through the corresponding slides below.
  • We recommend you to work through the Git and GitLabTM workflow tutorial to get basic knowledge of Git.
  • If you intend to work with sensitive personal data in the MaRC3 cluster, the training course on “Handling personal data in the HPC environment” must be successfully completed in advance. This is offered by the operators of the MaRC3 cluster and must be repeated every 2 years (reference to be announced).

Introduction to the MaRC3 HPC Cluster

You can log in to the cluster via commandline:
ssh -p 223 YOUR_STAFF_ACCOUNT@marc3a.hrz.uni-marburg.de
Use your Uni Marburg staff account and enter the respective password (without 2-FA) when prompted for it. Details are provided in the slides below.

The following slides from the “Introduction to the MaRC3 HPC Cluster” were kindly provided by René Sitt and the HKHLR.

© 2023 René Sitt. This work is openly licensed under CC-BY-SA 4.0

Ensure Git-LFS is enabled!

Be aware that purging modules may disable the "git-lfs" module. This may unintentionally cause large files to be handled by Git instead of Git-LFS which degrades performance and stability of your Git repo!

No heavy computation on login nodes

The login nodes are used by all users to prepare and queue actual compute jobs or interactive sessions. These will be executed on the compute nodes. Please do not perform computation tasks on the login nodes!

Owner group TAM

Members and collaborators to TAM get privileged access to that group's "owner" partition, which grants higher priority and relaxed max. runtime requirements on that group's subset of nodes.
To select this partition, either include the line #SBATCH --partition=owner_tam in your job scripts, or specify --partition=owner_tam directly in the job submit command.
Jobs that do not specify a partition will use the public partitions, whose defaults and limits are displayed at login.

Using SFTP client software

It is also possible to interact with the filesystems of MaRC3 with client software that uses SFTP protocol. Examples are FileZilla or WinSCP.