The ZIM operates a compute cluster for High Performance Computing (HPC) for the UP and supports members of the University of Potsdam (UP) in using the supercomputer system of the North German High Performance Computing Network (HLRN).
The UP-owned compute cluster is accessible under the hostname "orson2.rz.uni-potsdam.de". It is a cluster of
redundant head nodes,
The file systems are accessible from the head nodes and all compute nodes. The data on the scratch file systems are not backed up by us, so the users are responsible for backing up data that is needed for a longer period of time.
The distribution of the computing jobs to the nodes is done by the job and resource scheduler "SLURM". Direct user access to the computing nodes is not possible.
Currently, a restructuring of the storage is taking place. The data under /mnt/scratch and /mnt/largescratch is currently delivered by NFS and GlusterFS respectively. Both scratch directories will be merged and will be served by a BeeGFS. This will lead to a significant improvement in performance and thus an acceleration of IO-intensive jobs. More information about details will follow via the mail distribution list.
New Compute Nodes:
Ten new compute nodes with current hardware will be added to the cluster. Four of these are GPU nodes, each containing 3 Nvidia V100 GPUs. The GPU support enables significant acceleration for AI and machine learning applications, for example. The setup of these nodes will take place at the new location in the Institute of Computer Science, probably in April/May 2021.