Your benchmark may not guide a real application performance
summary
Argues that common microbenchmarks mislead because JS engines optimize hot paths: JetStream runs ~100 iterations and hits top speed after a few warmups, so microbenchmark gains often reflect JIT behavior, not real use. Shows real-case profiling of a video service where perceived slow startup hides a bootstrap/script execution pattern and an encoding conversion that made 59% of worst-case time go to painting. Recommends using profilers, reproducible harnesses, long-term dashboards and real user scenarios to find fixes that actually improve user-visible metrics.
AI-written summary · about
chapters (8)
- 0:00Introduction and setup
- 9:42Why performance matters
- 13:15Real scenarios over generic metrics
- 17:41Benchmarking pitfalls and JS engines
- 24:16Case study: Slow application startup
- 30:57Profiling tools and network impact
- 33:09Long-term benchmarking strategies
- 35:50Summary and key takeaways
related talks
02h 44mMeasure, don’t guess: benchmarking stories from the trenches
02h 41mMicro Benchmarking - The Art of Realizing One is Wrong
21mBenchmarking Like a Boss
27mAnd Benchmarks for All!
44mPractical guide to performance Benchmarks
38mJavaScript Performance Through a Spyglass
06mIGNITE: How to benchmark … poorly
35mBenchmarking: You're Doing It Wrong