Using containers for HPC workloads (but avoiding Docker)

Eric Burgueño

Containers are the old hot thing. But their adoption in all realms of IT is far from widespread. Most containers we see in the wild today are created to host daemon-type processes and scale them massively, but traditional HPC workloads are different and so HPC shops continue to have some aversion to them.

In this short talk I will discuss some of the challenges that Docker introduces when it comes to using containers in a multi-tenant HPC cluster, and what are some of the possible solutions. We will also explore the relationship between containers and reproducibility in Computational Science.