Cost-Effective Scheduling of a Massive Number of Containers in Kubernetes
summary
JD describes building Archimedes scheduler and a closed-loop stack including containerFS and ClickHouse monitoring to squeeze utilization on ~5,000 servers and ~225k containers, using ML-driven CPU/memory sizing, co-scheduling of mixed workloads, rack-aware placement and millisecond-level elastic scaling. Claim: handling an extreme June 18 sale with $24.7 billion transactions while cutting scheduling latency ~90% and lifting effective packing from ~40% to ~61%, plus memory-protection heuristics to avoid OOM kills.
AI-written summary · about
chapters (5)
- 0:00Evolution of containerization
- 3:51Customizing the Kubernetes
- 10:46Resource management strategies
- 18:09Architecting for efficiency
- 25:08Advanced scheduling results
related talks
38mCost-Effective Scheduling of a Massive Number of Containers in Kubernetes
11mHow Jenga & Tetris Can Help Run More Containers with Lesser Costs...
33mKubernetes cost optimization - How to actually do it- 11mLightning Talk: How Jenga & Tetris Can Help Run More Containers with Lesser Costs & Downtime
36mOptimizing Resource Usage in Kubernetes
24mKubernetes Batch Processing at Scale - A Scheduling Perspective
14mCut Costs, Not Performance: Implement Kubernetes Cost Optimization Strategies
43mContainers Change Everything