How to Think About Instrumentation Overhead
summary
Instrumentation adds measurable CPU, memory, bandwidth and sometimes startup and disk cost, and percent change follows 100 (instrumentation latency) / (original app latency) so a 2 ms instrumentation turns a 3 ms service into 5 ms (≈66% increase) but a 100 ms service into 102 ms (≈2%). Measure overhead with production‑like workload using tools like k6 load testing, control warm‑up and environment, avoid max‑stress tests, upgrade OpenTelemetry, apply sampling, disable unused modules and watch for manual instrumentation; Java agents currently contain about 100 instrumentations* and lack per‑instrument overhead breakdown.
AI-written summary · about
chapters (6)
- 0:00Introduction to instrumentation
- 1:25Understanding overhead
- 5:08The mathematics of latency
- 9:30Measuring overhead accurately
- 12:17Strategies for optimization
- 14:50Questions and answers
related talks
24mHow To Think About Instrumentation Overhead
32mWhat's the Performance Overhead? Answering the Biggest Question in Tracing
20mStrategies for Auto-Instrumenting Go Applications Without Code Changes
27mInstrumenting at 10 years per second
25mCalculating the Operations Cost of Software You Haven't Developed
31mDiscovering the Magic Behind OpenTelemetry Instrumentation
47mHow I Solved Production Issues with OpenTelemetry (And How You Can Too)
51mProduction Profiling: What, Why and How