cHPCe: Data Locality and Memory Bandwidth Contention-aware Containerized HPC

The High-Performance Computing (HPC) community started using Operating System virtualization, aka Containerization, due to its near-native performance compared to the BareMetal environment. Despite several existing container solutions for HPC, users are still unconvinced about the suitability of container orchestration solutions for their extreme-scale applications due to the lack of thorough performance assessment on recent advances. A key concern in current generation containerized HPC is the inherent performance degradation due to resource interference by co-hosted applications. This paper proposes an analytical model for data locality and memory bandwidth contention-aware container placement strategy for our developed containerized High-Performance Computing environment (cHPCe). Performance is evaluated and compared against LXD, Docker Swarm, Kubernetes, and Singularity using HPC Challenge and NAS parallel benchmarks. To the best of our knowledge, no study has been reported yet for such a comparative analysis with insight into data locality and memory bandwidth contention-aware container placement. The experimental result shows that data locality and memory bandwidth contention-awareness reduce the overall execution time of the benchmark in cHPCe by 51.41% in the best case compared to the Docker Swarm.