Delay accounting: an underrated feature of the Linux kernel
summary
Kernel exposes per-task wait-time counters (delay accounting) that measure how long a runnable task sits in the run queue, accessible via netlink socket to read per-task delay accounting or via schedstat counters. An open-source agent (Apache 2.0) already polls these values. Experiments show 100% CPU node load can either leave application latency unchanged or add roughly 2s of scheduling delay per second when a competing container has higher CPU shares, so delay accounting pinpoints true CPU starvation.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:23CPU scheduling basics
- 4:24Limitations of CPU metrics
- 8:47Delay accounting explained
- 11:48Real-world experiments
- 13:59Troubleshooting CPU delays
- 15:50Summary and Q&A
related talks
52mEffectively Measure and Reduce Kernel Latencies for Real-time Constraints
55mAdvanced Tracing Features to See What Your Application is Waiting On
50mrtla timerlat: Debugging Real-Time Linux Scheduling Latency
26mReal-time Linux: what is next?- 50mrtla timerlat: Debugging Real-Time Linux Scheduling Latency
32mCPU Power Saving Methods for Real-time Workloads
32mImprove CPU Utilization by Designing a New Schedule Algorithm
44mGetting Around to It: Deferred Work in the Linux Kernel