Optimizing Resource Usage in Kubernetes
summary
Adobe Experience Manager runs >70 clusters with 17,000 customer environments, ~200,000 deployments and ~18,000 namespaces and uses standard Kubernetes plus custom platform tooling to cut costs. Cluster autoscaler with a lease-aware bin packing reduces VM spend ~30–50%, switching to ARM images saves ~15–25%, Vertical Pod Autoscaler yields ~5–15% savings, and HPA on requests-per-minute cuts replica cost ~50–75%. Platform features include carpenter for right‑sizing nodes, KEDA for event-driven scale-to-zero, and JVM checkpointing like CRaC to turn 5‑minute starts into millisecond resumes.
AI-written summary · about
chapters (14)
- 0:00Optimizing Kubernetes resources
- 1:05Adobe Experience Manager use case
- 6:02Kubernetes requests and limits
- 13:16Cluster autoscaler and capacity
- 17:35Carpenter for node scaling
- 20:04Vertical Pod Autoscaler scaling
- 23:12Horizontal Pod Autoscaler strategies
- 26:25ARM workloads cost savings
- 27:30Hibernating environments
- 30:44Checkpointing containers
- 32:10KEDA event-driven autoscaling
- 33:58CO2 emissions optimization
- 35:26Avoiding API server overload
- 37:36Optimization strategies summary
related talks
36mOptimizing Resource Usage in Kubernetes
36mHow Adobe is Optimizing Resource Usage in Kubernetes
26mLessons Learned Migrating an Existing Product to a Multi Tenant Cloud Native Environment
15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.
14mCut Costs, Not Performance: Implement Kubernetes Cost Optimization Strategies
43mKubernetes cost optimisation - How to actually do it
25mThe Cloud Native Journey at Adobe
33mKubernetes cost optimization - How to actually do it