Kubernetes: Reducing Infrastructure cost and complexity
summary
Claims that pods drive infrastructure: Ocean by Spot provisions worker nodes to match pod CPU/memory/GPU/taints and predicts spot reclaim events to prefill capacity, claiming up to 90% cost reduction. Demo uses eksctl + EKS, metric-server, HPA, Elastigroup and Vegeta, drives 8,000 rps, HPA reaches ~100 replicas and Ocean brings five nodes to host 70–100 pods, then scales back to a single node; replacement vertical autoscaling and pod-level cost showback are implemented.
AI-written summary · about
chapters (9)
- 0:00Introduction to Kubernetes cost
- 3:14Kubernetes scaling challenges
- 9:11Pod-driven infrastructure approach
- 13:22Cost savings with instance types
- 16:56Ocean by Spot overview
- 21:28Live demo setup
- 22:52Demo scaling pods and nodes
- 31:59Demo results and summary
- 33:03Q&A and closing
related talks
43mKubernetes cost optimisation - How to actually do it
33mKubernetes cost optimization - How to actually do it
47mBuilding sustainable architectures with Kubernetes
14mCut Costs, Not Performance: Implement Kubernetes Cost Optimization Strategies
15mDemo: Resource(Cost) optimization on planet scale Kubernetes clusters.
26mDescubra las tendencias de optimización de costos en Kubernetes
36mCost-effective Compute Clusters with Spot and Pre-emptible Instances
44mFinOps at scale: Lessons from scaling Kubernetes workloads