Service overview

Kubernetes (K8s) is the recommended platform for submitting jobs to the ReCaS-Bari HPC/GPU cluster.

Compared with HTCondor, Kubernetes is the preferred choice when:
• your workloads are containerized; or
• your jobs are standalone, meaning they do not consist of hundreds of independent low-core tasks, and require GPUs and/or substantial computing resources.

Warning

Only containerized workloads can be executed on the cluster.

You may either use a third-party container image, use our default image, or build your own custom image.
Our default image is based on the configuration of our HTCondor worker nodes, making the transition to Kubernetes as transparent as possible for most use cases. This means that you can also submit jobs without any prior knowledge of containers by relying on the default image provided by the service.

For guidance on creating custom containers, please refer to this guide.

How to submit a request

Access to our HPC/GPU Kubernetes Cluster is available only for users with a ReCaS-Bari HTC/HPC account active. Users without such an account MUST request one.
Once your request has been submitted, it must be approved by the datacenter's technical coordinators. This process usually takes a few working days. You will receive an automatic email notification as soon as your account is activated.

After that, you can request access to the Kubernetes service using this link.

Please provide the following information:

   Title: “ReCaS HPC/GPU: request to access HPC/GPU K8s cluster”  
   Issue:
   - Name and Surname
   - HTC/HPC service username

You will be notified by email once access has been granted.