The International Conference for High Performance Computing, Networking, Storage, and Analysis
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Workshop:HPCSysPros21 Workshop
Authors: Samuel Knight (Sandia National Laboratories)
Abstract: HPC software stacks require administrators to deploy and maintain a complex collection of software. Historically this was done on bare-metal nodes, which present challenges when maintaining a coherent suite of software. This paper describes hosting administrative HPC software on a Kubernetes platform, along with auxilary tooling for storage, IP/DNS management, metrics and logging. Since Kubernetes exclusively runs container images, individual components of the software stack run in simplified, reproducible environments. Altogether, the platform can replicate most of the advantages of a bare-metal environment, while improving on reproducibility, stability, uptime, and the quality of telemetry.
Back to HPCSysPros21 Workshop Archive Listing