Enabling OpenMPI workloads on bare-metal infrastructure using Kubernetes

Abstract

The purpose of this bachelor’s thesis is to design and develop a software architecture for HPC clusters to support MPI-based workloads. We will demonstrate how many abstractions such a complex solution involves and how the cluster works from the software perspective. As a result, we will configure a two-node cluster, develop software architecture for it, and run workloads on top of it. We will use many technologies to develop our solution. To better understand the architecture, we will give a background on the most complex parts like Kubernetes and Docker.