Optimizing Kubernetes for Low-Latency and Real-Time Containers
summary
Shows how NRI + balloons plugin in containerd and CRI‑O actually reconfigures OCI metadata, pins CPU sets, allocates CPU/memory pools, moves IRQs, and toggles frequency and C‑states to cut 99th‑percentile wake‑up jitter (measurements on a Xeon P‑core node report up to ~98% reduction). Uses real‑time scheduler classes, per‑pool frequency/C‑state controls and daemonset deployment; works without modifying images, but node privileges are required and some knobs don’t apply to virtual nodes.
AI-written summary · about
chapters (7)
- 0:00Introduction to low-latency Kubernetes
- 3:38Problem with default Kubernetes setup
- 8:19Node Resource Interface explained
- 12:16Balloon plugin for resource control
- 20:06Performance results with balloons
- 25:49Summary and NRI benefits
- 28:42Q&A session
related talks
21mDo Not Disturb Mode: Fine-Tuning Resources for Latency Sensitive Workloads
36mMitigating Noisy Neighbours: Advanced Container Resource Management
25mTune It Up! Enabling Low Latency in Kubernetes Clusters
33mCRI-O Still Loves Kubernetes
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
31mCRI-O: The Runtime Control Room
32mOptimizing Application Performance on Kubernetes
40mNode-Level Optimization in Large Cluster Powered by Containerd Plugin