Scale and Performance Testing of Kubernetes: Answers for Specific Applications
summary
A 500 physical‑node lab simulates up to 5,000 minions and runs an OpenStack microservice deployment to measure real behavior: scheduling time scales roughly linearly but shows plateaus at ~500 pods that create long gaps and unpredictable grouping of ReplicaSets, rolling updates complete predictably when kubelet poll settings are tuned (about 20 minutes for a 500‑pod ReplicaSet with aggressive intervals), Calico gets close to wire‑speed (~90%) networking, while docker hangs, kube‑dns overload and difficulty troubleshooting are practical failure modes.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:05Mirantis Performance Team
- 3:23Lab Infrastructure: HW
- 4:17Lab Infrastructure: Networking
- 5:17Lab Infrastructure: Automation
- 12:56OpenStack On Kubernetes
- 15:09OpenStack Service Details
- 16:43OpenStack Big Picture
- 18:38Basic Tests: ReplicaSet deployment
- 21:22Basic Tests: ReplicaSet creation process
- 24:04Basic Test: ReplicaSet rolling upgrade
- 25:46Basic Test: ReplicaSet with nodeSelector
- 26:47Basic Test: Multiple RS in a single yaml
- 29:30Application Test: OpenStack Deployment
- 31:38Calico Network Performance
- 32:43Performance Summary
related talks
34mPanel: Improving and Managing Kubernetes at Scale
31mLoad Testing Kubernetes: How to Optimize Your Cluster Resource Allocation in Production [I]
36mUnderstanding Scalability and Performance in the Kubernetes Master
27mThe Kubernetes Hunger Games: Distro Performance in the Edge
32mScaling Kubernetes Application to 1M Users
35mKubernetes, Lessons Learnt
02h 26mKubernetes Performance Tuning Workshop
29mScale Your Batch / Big Data / AI Workloads Beyond the K8s...