Migration plan

We will migrate Hábrók to Alma Linux 9 in a three step procedure. The first step consists of a testing phase, in which some nodes will be made available to the users for testing their workloads on Alma Linux 9. This is followed by a migration phase, in which compute and user interface nodes will be gradually moved over from Alma Linux 8 to 9. The final phase is when all nodes have been migrated, including the last user interface nodes.

We have made a few nodes available, which are running Alma Linux 9, to allow users to test their software and workloads. The following nodes are available:

1. Interactive login node merel1.hb.hpc.rug.nl. Users can login to this node and use it in the same way as the other interactive CPU based nodes.
2. Two regular compute nodes called vink1 and vink2. These nodes are CPU based and have 128 cores and 512 GB of memory.
3. A V100 GPU node called groenling1 with 2 Nvidia V100 GPUs.

These nodes have been put in a special reservation called alma9_testing. Jobs can be submitted to this reservation using the following sbatch option: -J alma9_testing. This can be given on the command line, or using the following #SBATCH line in your job script:

#SBATCH -J alma9_testing

As you can see from the host names the compute nodes will be named after common birds in the Netherlands, using their Dutch names.

We plan to run this testing phase for three weeks. Please report any issues you encounter while testing the new Alma 9 nodes, so that we can find a way on how to fix these.

After the testing phase we will start the migration of the rest of the infrastructure to Alma Linux 9. This includes the following steps:

1. We will remove the reservation from the existing test nodes and these will become available for all jobs, that do not explicitly request Alma Linux 8 (see below for details on this)
2. We will migrate login2.hb.hpc.rug.nl, interactive2.hb.hpc.rug.nl and gpu2.hb.hpc.rug.nl to Alma Linux 9.
3. We will migrate the other Hábrók compute nodes to Alma Linux 9 in batches. Nodes will only be migrated when the running jobs have finished.

If you want to make sure your jobs land on a node with specifically Alma Linux 8 or 9, you can put a constraint in your jobs using the sbatch parameter --constraint. You can either specify --constraint=alma8 or --constraint=alma9.

In a job script this would look like:

#SBATCH --constraint=alma8

Or something similar with alma9.

The last steps of the migration will that the last compute nodes will be migrated and the migration will be completed by reinstalling login1.hb.hpc.rug.nl, interactive1.hb.hpc.rug.nl and gpu1.hb.hpc.rug.nl.

After this point all user interface and compute nodes are running Alma Linux 9.