Containerization of parallel MPI and accelerated applications on HPC systems

Using containers in Linux has become a common practice in many business environments. Technologies like Docker or Kubernetes, allow to create self-contained environments or ''containers'' that are designed for a specific purpose. Unfortunately, the use of these technologies in High Performance Computing (HPC) environment has been problematic due to security concerns. The open source project Singularity addresses this problem, allowing for the creation of containers that maintain the security infrastructure of the machine they are running on. After a general introduction of Singularity, two containerized applications are presented: 1) Quantum Espresso, the integrated suite of Open-Source computer codes for electronic-structure calculations and materials modelling at the nanoscale, and 2) Tensorflow, the open-souce software library for machine learning. The performance comparison between bare metal and container executions is provided, showing a negligible overhead introduction in the containerized version of the applications. All the tests have been run on the supercomputing clusters installed in CINECA.

Comments

Comments are closed.