A Toolkit for Simulating Kubernetes Scheduling at Scale
Talkat KubeCon + CloudNativeCon North America 2019 · on Wednesday, 2019-11-20· at 16:25· Room 15AB - San Diego Convention Center, Mezzanine Level·
summary
A simulator runs a real Kubernetes control plane and emulates thousands of kubelets inside containers to simulate 10K+ nodes and ~200K pods on a single machine, ingesting real cluster events or synthetic workloads to replay real event traces. Implementation uses telegraf + InfluxDB, exposes a dashboard with per-stage scheduler latencies and placement heatmaps, and drives concrete results at JD.com — e.g. improved scheduler QPS from ~2 to ~40 and 61% average memory consolidation.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:35JD Retail Technical Infrastructure Group
- 3:05JD Container Platform
- 6:06Cluster Simulation Tools
- 7:08Key Requirements
- 7:57Outline
- 8:46JoySim: A Kubernetes Cluster Simulator
- 11:07Architecture
- 12:38Events
- 15:13Node Simulation
- 16:36Implementation and Deployment
- 17:42Monitoring and Visualization
- 19:48Case Study: APIServer Evaluation and Optimization
- 23:35Case Study: Scheduler Performance Optimization Simulation testbed
- 26:46Case Study: Online Scheduling
- 29:27Summary
- 30:31Contact
related talks
27mUnderstanding Kube-Scheduler-Simulator
25mKube-scheduler-evaluator: Evaluating Kubernetes Schedulers at Hyperscaler…
37mA Scheduling Simulator for Capacity Estimation of Kubernetes Clusters
01h 29mTutorial: Unleash the Full Potential Of Kubernetes Scheduler: Configuration, Extension And Operation In Production
25mScale your Batch / Big Data / AI Workloads Beyond the Kubernetes Scheduler
48mScale and Performance Testing of Kubernetes: Answers for Specific Applications
29mScale Your Batch / Big Data / AI Workloads Beyond the K8s...
32mBuilding a Kubernetes Scheduler using Custom Metrics