Performance and Scalability Tuning Kubernetes for OpenShift and Docker
summary
Built a reproducible OpenShift-on-OpenStack environment in the CNCF lab and drives it to failure to find fixes and guidance, using pre-baked images, NVMe, Ansible tuning and distributed tracing with pbench. Runs a 1000-node cluster test that creates 13,000 projects and ~52,000 pods, files 30+ bugs (including an API server crash), and publishes test code (cluster-loader) and capacity formulas so deployments size disk, CPU and SDN correctly.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 4:18Food Groups of Performance
- 8:25Container Image Format
- 11:49CNCF Cluster
- 15:59ETE Test Harness
- 16:59Performance at Scale
- 23:14Takeaways
- 24:04Capacity Planning
- 25:34Ansible
- 26:49Questions
- 34:29CPU bound workloads
- 40:44API response latency
related talks
28mOpenShift for the Enterprise or What Could Possibly Go Wrong
30mTuning and automating for Telco 5G containerized workloads
47mInfrastructure under Test in OpenShift
22mOpenShift, or Kubernetes on steroids
31mLoad Testing Kubernetes: How to Optimize Your Cluster Resource Allocation in Production [I]
30mK8s Under Fire: Igniting Performance & Burning Through Bottlenecks with Kube-Burner
13mOptimizing Workload Management in a Multi-Cluster Environment: A Case Study
50mOpenShift in Production: 7 Pitfalls to Avoid