How Workload Rightsizing Increases Availability
summary
Storm Forge deploys Optimize Live to train a per-container ML model on metrics (CPU, memory, requests, limits) with a 28‑day lookback, auto-generating k8s patches that adjust requests/limits and HPA targets to avoid thrash. Rolled across 10,000 tenants (30k namespaces, 300k+ containers, ~1.2M resource fields), it moves changes from weekly toward twice‑daily, reports ~40% cluster node reduction in examples, speeds migration from two years to six months, and lifts aggregate availability to 4+ nines.
AI-written summary · about
chapters (12)
- 0:00Introduction and speaker intro
- 1:47Speaker background and team
- 3:08Customer case study overview
- 5:16Resource request challenges
- 8:03Automating right sizing at scale
- 9:08Right sizing fundamentals
- 11:14Machine learning approach
- 13:21Key use cases for ML
- 16:55Implementation and rollout
- 19:44Outcomes and availability gains
- 22:11Reduction in churn and efficiency
- 24:02Q&A and resources
related talks
22mHow to make Dev and Ops folks look good as an SRE
40mOptimizing AI/ML Workloads
19mResource Management Innovation for AI Workloads in Cloud Computing
25mTackling operational and capacity utilization challenges at a large scale enterprise deployment
31mFailing to Autoscale?
27mLeverage ML to automatically tune K8s apps
18mLet the machines optimize the machines
10mLightning Talk: Does Cloud Elasticity Pay off for HPC Workloads?