Skip to main content

Calypso software stack

The software configuration of the Calypso students cluster.

User accounts​

User access is SSH based, managed by the staff — see Getting access.

Container runtimes​

To run containers on Calypso, you can use:

SLURM cluster​

There is a SLURM cluster on all Calypso worker nodes:

NodeRole
calypsomasterno SLURM
calypso0SLURM controller + accounting DB
calypso[1-7]SLURM workers
This SLURM install is stale and is to be rebuilt

Treat the details above as historical: the cluster is hand-maintained (slurm.conf is a configurator.html file copied to each node by hand, not templated) and is slated for reinstallation rather than repair. calypso8 was commented out of slurm.conf on 2026-08-04 when it became carnaval0; the powered-off nodes still hold older copies, which the rebuild makes moot.

Configuration​

GPU resources​

Each node in the cluster has the NVIDIA Container Toolkit installed, which allows you to use their NVIDIA GPUs from within containers:

  • calypso[0-9] are equipped with NVIDIA A2 GPUs
  • calypso[10-14] are equipped with NVIDIA Tesla T4 GPUs

Docker registry​

To avoid saturating the network link with builds, there is a local registry that can be used on calypsomaster. The daemon.json of each node is already configured with this local insecure registry:

{
  "insecure-registries": ["192.168.88.248:5000"]
}

The registry is effectively dead (the container runs with no published ports — verified 2026-08-02) and is slated for retirement with the calypsomaster conversion; the earlier plan of re-creating it behind a local PKI is superseded by that conversion.