Rocky 9 migration#

The KU Leuven Tier-2 cluster wICE will migrate to Rocky 9.5 as the operating system. Important differences at the system level are listed below.

Packages

Rocky Linux 8

Rocky Linux 9

kernel

4.18.0-513.24.1

5.14.0-503.38.1

bash

4.4.20

5.1.8

gcc

8.5.0

11.5.0

glibc

2.28

2.34

Centrally installed modules have already been made available for Rocky 9, starting from toolchain 2021a.

Timing#

The wICE nodes will be migrated to the new OS in November 2025. We foresee a test period for all users until the end of October. The Genius cluster will be decommissioned with the arrival of a new Tier-2 cluster at the end of the year. Genius will not be migrated to Rocky 9 anymore. The new Tier-2 cluster will be launched with Rocky 9.

Please use the test period to try out the new OS before the actual migration in order to avoid interrupting your workflow.

Note

Be aware that toolchains older than 2021a will no longer be available after Genius has been decommissioned.

As always you can contact hpcinfo@kuleuven.be in case you have questions or remarks.

How to test#

We have reserved several nodes of the batch, batch_sapphirerapids, gpu_a100 and gpu_h100 partitions for testing purposes. To send jobs to these test nodes with Rocky 9, simply submit your job with the --reservation=rocky9_pilot option, for example:

$ sbatch --reservation=rocky9_pilot myjobscript.slurm

Partition

CPU/GPU type

Number of nodes

batch,batch_icelake

IceLake

12

batch_sapphirerapids

Sapphire Rapids

12

gpu_a100

IceLake / A100

1

gpu_h100

Genoa / H100

1

As the pilot phase continues we may add more nodes to this reservation. The following query can be used to look up the currently reserved nodes:

$ scontrol -M wice show reservation rocky9_pilot --json | jq ".reservations[0].node_list"

If you are not a vsc3* user and also want to test, please ask the helpdesk to add you to the reservation.

Expected impact#

The impact of this upgrade will be small for most users. If you are only using centrally installed modules, your module load commands will automatically load the appropriate modules (e.g. the ones installed for Rocky 9 if you are on a node with Rocky 9). Note that this may not apply if you are manually modifying your module path (if in doubt, please consult The module system on Leuven clusters).

Note

If you have been compiling your own software on Rocky 8, it is possible that this software will not run on Rocky 9. If this is the case or if you have any doubts, we recommend to recompile on a node with the new OS. When doing so, it can be convenient to use the ${VSC_OS_LOCAL} variable which describes the node’s operating system (i.e. “rocky8” or “rocky9”).

Note

Keep in mind that also Python or R package installations may involve compiling steps for extensions and so may need to be redone for Rocky 9.

Note

Conda environments created on Rocky 8 will normally continue to work on Rocky 9 (at least if the compiled components are provided by Conda packages, as is normally the case).

Known issues#

  • Currently the CPU cores are unable to reach the maximal (‘turbo’) frequency. Compared to nodes with Rocky 8, you may therefore see somewhat lower performance if only a few cores are active while the other cores are idling. This issue is still being investigated.