How To Run Kubernetes Pods On My Slurm-Based HPC Center
summary
A practical deploy guide for interLink that maps plain Kubernetes pods onto Slurm by translating pod manifests into batch jobs: it uses three core components (a virtual kubelet provider, an interLink server on the HPC edge, and an optional network plugin), requires a recent Kubernetes, container runtime, shared filesystem and unshare capability, and supports wildcard ingress for full features. Demo runs on the Leonardo supercomputer showing Kubeflow notebooks launched as normal pods that create Slurm jobs, expose logs, and provide GPU access.
AI-written summary · about
related talks
32mPods Everywhere! InterLink: A Virtual Kubelet Abstraction Streamlining HPC Resource Exploitation
39mSlinky: Slurm in Kubernetes, Performant AI and HPC Workload Management in Kubernetes
30mThousands of Virtual Kubelets: 1-to-1 Mapping a Supercomputer To Kubernetes With Supernetes
12mLightning Talk: {”spec”: “nodeSelector”: "EuroHPC"}
41mBuilding & Operating a Large-scale HPC AI Cluster on Kubernetes
46mBuilding Supercomputers Using Kubernetes and OpenHPC
28mKubeFlux: An HPC Scheduler Plugin for Kubernetes
08mLightning Talk: Orchestrating Kubernetes Clusters on HPC Infrastructure