Evaluation of Containers for HPC

Some of the main challenges in scientific computing today deal with performance-preserving portability of software and reproducibility of the final results; likewise, with the advent of modern cloud computing, these, along with other issues like deployment, are also found in the sider software and computing world. Containers can help solve all of these issues by packing the software along with its dependencies together, in an easy-to-distribute and lightweight format. Herein we investigate the utility of Singularity, a HPC targeted container solution which overcomes the main issues with deploying more mainstream solutions such as Docker. Singularity is found to be both suitable and easy to deploy with the current set-up at CERN. The performance costs are minimal in accordance with the previously reported figures for Singularity and does indeed behave well when submitted through Slurm. This report also considers the new possible extensions to the software-infrastructre enabled by containers that can run on several different systems without any additional compilation or configuration. Lastly, specific use-cases such as Fire Dynamics Simulator and Warp are containerised and deployed to the users.