Optimizing Resource Usage in Kubernetes
summary
Runs a cloud service on more than 50 clusters with ~17,000 customer environments, ~200,000 Kubernetes deployments and >10,000 namespaces; uses cluster autoscaler (30–50% cost reduction), HPA, VPA (k8s v1.27 caveats), multi-arch images (≈20% savings on ARM), and Argo Rollouts for canaries. Builds a homegrown hibernation to zero for sandboxes (scale-to-zero after 4 hours, ~85% hibernated) and recommends KEDA/Knative for event-driven scaling, immutable ConfigMaps/Secrets to reduce etcd load, and watching pod-per-node limits to avoid wasted capacity.
AI-written summary · about
chapters (13)
- 0:00Kubernetes resource optimization
- 1:11Adobe Experience Manager use case
- 5:52Kubernetes requests and limits
- 9:48CPU throttling in Java apps
- 13:04Cluster autoscaler strategies
- 16:15VPA limitations and use cases
- 19:08HPA challenges and solutions
- 22:46VPA vs HPA for scaling
- 23:50ARM architecture benefits
- 24:55Hibernation for dev scaling
- 27:49CO2 emissions in cloud regions
- 30:00Control plane resource limits
- 32:47Optimization summary
related talks
39mOptimizing Resource Usage in Kubernetes
15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.
28mScaling in Kubernetes Safely on On-Prem KaaS Across 1,300+ Clusters and 40,000+ Nodes
14mCut Costs, Not Performance: Implement Kubernetes Cost Optimization Strategies
11mCost and Resource Optimization In Kubernetes
43mKubernetes cost optimisation - How to actually do it
47mPostmortem: A Modern Approach to Managing Kubernetes Clusters
27mMaximizing Developer Productivity in Kubernetes