Delay Accounting: An Underrated Feature of the Kernel

Is your application actually getting the CPU time it needs, and how would you know if it wasn't? This session explores delay accounting, a built-in Linux kernel feature that most engineers overlook, and what it reveals about CPU time shortages and application latency that traditional metrics cannot.

Checking CPU usage or load average gives you a surface-level picture, but it doesn't tell you whether individual processes are being starved of kernel resources. The Linux kernel has precise, built-in mechanisms for measuring exactly how long each task waits, and this talk shows you how to put them to use.

We dive into how delay accounting works and how to leverage these kernel-level metrics for more accurate performance analysis and system optimization. If you work on infrastructure performance and want deeper visibility into what's actually happening beneath your applications, this session offers a practical look at an underused but powerful observability tool.