Optimizing Application Performance on Kubernetes
summary
Starts with observability using Prometheus and Grafana and recommends exportable app metrics, BIOS/OS tuning (scaling governor change can cut performance by up to 30%), and matching pods to nodes with node affinity and pod anti‑affinity. Introduces Cruise autotune that uses Bayesian optimization over ~100 trials to tune container CPU/memory and JVM/Quarkus flags; on a Quarkus REST benchmark it finds a config that yields ~83% improvement in median response time and about an 8% cost reduction.
AI-written summary · about
chapters (14)
- 0:00Start
- 2:26Lower My Response Time!
- 3:16Observability
- 4:43Don't Forget The Hardware
- 6:15Node and Pod Affinities Node Affinity Helps to match workloads to right resources Helps to schedule related pods together
- 10:14It's All About the Scaling
- 12:02Optimize the App Stack?
- 12:52Life of a SRE?!
- 13:35Autotune Architecture
- 19:39Objective Fn: Reduce Response Time
- 21:58Autotune vs Default Config - Take 3
- 24:48Autotune vs Default Config - COST
- 25:18Best Configuration
- 25:37Summary
related talks
15mOptimizing Application Performance on Kubernetes
27mMastering Cloud Cost Optimization with Kruize
29mRight-Size Your Kubernetes Deployments
02h 26mKubernetes Performance Tuning Workshop
11mK8s Resources Ensuring Performance and Optimization
26mDescubra las tendencias de optimización de costos en Kubernetes
25mTune It Up! Enabling Low Latency in Kubernetes Clusters
46mKubernetes. From 0 to Production-Grade with Java.