Urblauna
Kesako?
Urblauna (Romanche), or Lagopède Alpin in French, is a bird known for its changing plumage which functions as a very effective camouflage. More information is available at https://www.vogelwarte.ch/fr/oiseaux/les-oiseaux-de-suisse/lagopede-alpin
It's also the name of our new sensitive data compute cluster which will replace the Jura cluster.
Information on how to connect to Urblauna can be found here.
Hardware
Compute
The cluster is composed of 18 compute nodes of which two have GPUs. All have the same 24 core processor.
| Number of nodes | Memory | CPU | GPU |
| 16 | 1024 GB | 2 x AMD Epyc3 7443 | - |
| 2 | 1024 GB | 2 x AMD Epyc3 7443 | 2 x NVIDIA A100 |
The GPUs are partitioned to create 4 GPUs on each machine with 20GB of memory per GPU
Storage
The storage is based on IBM Spectrum Scale / Lenovo DSS and provides 1PB of space in the /data filesystem.
Whilst reliable this space is not backed up and all important data should also be stored on /archive
The Curnagl /work filesystem is visible in read-only mode on Urblauna and can be used to install software on an internet connected system before using it on Curnagl.
|
Filesystem mount point |
Description |
| /users | Urblauna home directory |
| /scratch | Urblauna scratch space |
| /data | Urblauna data space (no backup) |
| /archive | Secure data space with backup (login node access only) |
| /work | Curnagl data space (read only) |
| /jura_home | Jura home directories (read only, login node only) |
| /jura_data | Jura data space (read only, login node only) |
Software
For information on the DCSR software stack see the following link:
https://wiki.unil.ch/ci/books/high-performance-computing-hpc/page/dcsr-software-stack
Slurm partitions
On Urblauna there are two partitions - "urblauna" and "interactive"
$ sinfo
PARTITION AVAIL TIMELIMIT NODES STATE NODELIST
urblauna* up 3-00:00:00 17 idle sna[002-016],snagpu[001-002]
interactive up 8:00:00 4 idle sna[015-016],snagpu[001-002]There is no separate GPU partition so to use a GPU simply request
#SBATCH --gres=gpu:1
To launch an interactive session you can use Sinteractive as on Curnagl