How Much Is Tracing? Measuring the Overhead Caused by the Tracing Infrastructure
summary
Measures GPIO toggles on a BeagleBone Black with an oscilloscope and a tight-loop kernel driver and reports concrete overheads for tracing options. A real-time kernel yields ~145 ns pulses and enabling tracing jumps to 168 ns (+23 ns). Simple trace events add about 1.36 µs, synthetic events another ~0.4 µs, kernel probes ~2.9 µs, eBPF ~3.7 µs, function tracer shows a ~3 µs base with function graph reporting ~16 µs vs scope 12 µs. Raw data and scripts sit in the cloud repository.
AI-written summary · about
chapters (4)
- 0:00Start
- 26:56Synthetic event on his trigger
- 38:02Reference for userspace measurement
- 40:11Trace marker
related talks
52mTracking Down What Happened: A Comparison of Tracing Solutions
34mTracing: The Bane of You Security Folks
35mLinux Tracing updates
05mPerformance analysis using Systrace
27mLow-Overhead Tracing Using eBPF for Observability into Kubernetes Apps and Services
41mSynthetic Events and Tracing Latency Within the Kernel
46mbpftrace internals
25mKernel tracing with BPF