Rootless Kubernetes on HPC: Early Experience and Future Outlook¶
Kamil Tokmakov (NEC Deutschland)
Abstract¶
Recent efforts to bring cloud-native software stacks to HPC environments - particularly for data scientists and ML/AI user groups - have focused on supporting containerized workloads and provisioning Kubernetes clusters in various integration configurations. These include integrating HPC job schedulers on top of and alongside Kubernetes (e.g., via the slurm-operator and slurm-bridge components of the Slinky project), as well as deploying Kubernetes clusters as part of individual batch jobs. While the first two approaches require significant modifications to traditional HPC infrastructure and involve increased operational complexity, the latter configuration relies on rootless containers and is particularly useful for supporting the cloud-native users with a familiar software stack without compromising HPC users and infrastructure. In this talk we share our early experience with managing rootless Kubernetes on HPC.