OpenShift for low latency and real time workloads
summary
End-to-end OpenShift recipe for hitting a 20 microsecond OS latency budget for 5G telco DU workloads, using Performance Addons Operator, Tuning Operator and MachineConfigOperator to install a real-time kernel, reserve/ isolate CPU pools, and configure Kubernetes cpu manager/topology manager and a new runtime class for DPDK. Validation uses hardware latdetect, cyclictest and DPDK-like load generators; typical runs are 12–24 hours and one debugging result shows three 120 µs spikes in ~100 million samples. Known caveats: isolated CPUs still count toward capacity and guaranteed pods may wait ~5–10 seconds to receive exclusive CPUs.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:08What is low latency
- 2:55Time constraints
- 4:11Low latency tuning
- 6:27Openshift role
- 7:40Kubernetes role
- 9:16Performance profile
- 12:12Workload pod
- 13:58Latency measurement tools
- 15:19Latency tuning validation
- 16:45Common issues
- 18:17Tips
related talks
30mTuning and automating for Telco 5G containerized workloads
25mTune It Up! Enabling Low Latency in Kubernetes Clusters
44mPerformance and Scalability Tuning Kubernetes for OpenShift and Docker
36mRunning High-performance User-space Packet Processing Apps in Kubernetes
31mRunning Low-Latency Workloads on Kubernetes
35mMigrating the Next Generation Mobile Core Towards 5G with Kubernetes
27mCrafting Seamless Development: Improving Kubernetes Operator Developer Experience
02h 26mKubernetes Performance Tuning Workshop