Abstract: Containers have a considerable advantage for application portability in different environments by isolating process with a small performance overhead; thus it has been rapidly getting popular in a wide range of science fields. However, there are problems in container image configuration when run in multiple HPC environments, and it requires users to have knowledge of systems, container runtimes, container image format, and library compatibilities in HPC environments.
In this study, we introduce our HPC container workflow in multiple supercomputing environments that have different system/library specifications (ABCI, TSUBAME3.0). Our workflow provides custom container image configurations for HPC environments by taking into account differences in container runtime, container image, and library compatibility between the host and inside of the container. We also show the parallel performance of our application in each HPC environment.
Best Poster Finalist (BP): no
Poster summary: PDF
Back to Poster Archive Listing