Towards Sustainable Multi-Cluster Management in Real-World Academic Infrastructure
summary
Lab runs four bare-metal clusters over 100GbE with 2PB Rook-Ceph storage and a 22‑GPU pool (7 accelerator types including L40S and A100) unified as ScaleX‑POD. Management uses SmartX repos plus OpenKISS PXE provisioning, ArgoCD for apps, Karmada for placement and Kyverno for PSS Restricted and default‑deny NetworkPolicies with namespace RBAC. Migration hits: no spare capacity so a 10‑node landing zone was built; ArgoCD ownership handoff pruned running apps and stalled until autosync and cascading deletes are sequenced; solution favors rollouts to new clusters and PR‑pinned releases.
AI-written summary · about
chapters (8)
- 0:00Sustainable multi-cluster intro
- 1:09Problem in research infra
- 5:07Design for sustainability
- 17:06Migration challenges learned
- 22:30SmartX cluster setup demo
- 25:11App deployment with SmartX Scalex
- 27:23Multi-cluster monitoring demo
- 28:30Q&A
related talks
08mLightning Talk: Orchestrating Kubernetes Clusters on HPC Infrastructure
19mUsing Rook in Production on National Research Platform
31mManaged Kubernetes — Next Gen Academic Infrastructure?
35mBest Practices of Production-Grade Rook/Ceph Cluster
23mImplementation Challenges: From HPC to Containers in the Academy
36m200 Clusters vs 1 Admin
34mInfrastructure at Scale Using ArgoCD
01h 21mHands On with Rook: Ceph & Kubernetes