Performance Effects Of Running Container-Based Open-Mpi Cluster In Public Cloud

The vast majority of HPC users are heavily leveraging MPI middleware for their applications needs. For almost two decades, providing an infrastructure that constantly answers the increasing demand, security standards, software interoperability, and the ineffective resources utilization are the challenges placing HPC administrators between the overhead of the virtualization and the manual tuning of the performance pick, forcing them to find the edge of the consensus on their own.Recently, developments like Linux Containers, Infrastructure as Code, Public Cloud opened a new horizon in front of the industry, by redefining the engineering roles and providing tools and practices for solving some major caveats from the past.This paper presents three architectures for setting up Open-MPI cluster in a Linux Container-Based environment and explains how these architectures can be implemented across private and public cloud with the main focus of the performance effects in the public cloud.