SC19 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Poster 88: HPC Container Runtime Performance Overhead: At First Order, There Is None

Authors: Alfred Torrez (Los Alamos National Laboratory), Reid Priedhorsky (Los Alamos National Laboratory), Timothy Randles (Los Alamos National Laboratory)

Abstract: Linux containers are an increasingly popular method used by HPC centers to meet increasing demand for greater software flexibility. A common concern is that containers may introduce application performance overhead. Prior work has not tested a broad set of HPC container technologies on a broad set of benchmarks. This poster addresses the gap by comparing performance of the three HPC container implementations (Charliecloud, Shifter, and Singularity) and bare metal on multiple dimensions using industry-standard benchmarks.

We found no meaningful performance differences between the four environments with the possible exception of modest variation in memory usage, which is broadly consistent with prior results. This result suggests that HPC users should feel free to containerize their applications without concern about performance degradation, regardless of the container technology used. It is an encouraging development on the path towards greater adoption of user-defined software stacks to increase the flexibility of HPC.

Best Poster Finalist (BP): no

Poster: PDF
Poster summary: PDF

Back to Poster Archive Listing