Score-Driven Multi-Cluster Management: An Evaluation Framework for…
summary
They build a new Open Cluster Management add-on called Dynamic Scoring Framework (DSF) that runs lightweight agents on managed clusters, pulls Prometheus metrics, and emits compact add-on placement score scalars to the hub instead of shipping raw telemetry. An external optimizer reads those live performance and power scores, then creates or updates OCM Placements and changes MIG partitioning and replica counts. Demo shows automatic cluster switching and GPU partition changes when goals flip between throughput and power, and the system successfully drives placement decisions in real time.
AI-written summary · about
chapters (9)
- 0:00Multi-cluster management intro
- 1:48AI workloads need multi-cluster
- 4:56Open Cluster Management foundation
- 9:31Dynamic scoring framework
- 13:48Optimization with live scores
- 15:41Demo: score-driven placement
- 20:21Multi-cluster optimization pattern
- 21:46Key takeaways and future
- 22:50Q&A on scoring
related talks
05mScheduling at the Edge of Reason: Multi-Cluster AI & OCM
05mProject Lightning Talk: Scheduling AI Workload Among Multiple Clusters
01h 12mBuilding scalable and flexible cluster managers using declarative programming
35mAcing Distributed Container Management
41mIntroducing
34mWorking your Cluster: Smarter Scheduling Decisions for Your Work...
29mMulticluster Kubernetes Management Made Easy With Open Cluster Management
28mScale Smarter Not Harder: How Extending Cluster Autoscaler Saves Millions