Optimizing Application Performance on Kubernetes
summary
Shows that CPU MHz swings (3.5 GHz peak dropping to less than half) and power‑save BIOS governors can cut performance ~30%, so tune BIOS and OS scaling governors and ignore hyperthreading for capacity (HT gives ~20% at best). Emphasizes observability tradeoffs—per‑second metrics vs storage/CPU overhead—and concrete actions: set requests to cover week‑long observed peaks, limits for spikes, use HPA on app metrics, VPA is black box, and use Kruize + Spring actuator/IBM Health Center for JVM heap and RAM‑% recommendations with Grafana dashboards and API.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:40Performance tuning knobs
- 3:07CPU scaling governors
- 4:53Hyperthreading capacity
- 5:16Observability in Kubernetes
- 7:08Right-sizing application pods
- 9:11Vertical pod autoscaling
- 11:01Introducing Cruz tool
- 14:13Summary and Q&A
related talks
32mOptimizing Application Performance on Kubernetes
47mSecrets of Performance Tuning Java on Kubernetes
31mRunning Low-Latency Workloads on Kubernetes
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
35mSustainability Research the Cloud Native Way
02h 26mKubernetes Performance Tuning Workshop
33mUncover Kubernetes cost-optimization trends
51mWhy You Need Performance Tests for Proper Kubernetes Scaling