About this tool
What does it build?
Structured benchmark harnesses: define operations, iteration counts and warmups, then measure throughput and latency distributions instead of guessing which code is faster.
How to use it
- Describe the operations to compare.
- Set warmup + measured iterations.
- Read median/p99, not just the mean.
Examples
- JSON.parse vs manual parsing on your payload shape.
- Regex vs string split for a hot log path.
- Comparing sort implementations on realistic data sizes.
Important limitations
- Microbenchmarks lie easily: JIT warmup, GC pauses and CPU throttling distort small differences.
- Measure your real workload shape — synthetic winners often lose in production.
- One machine, one result: hardware and runtime versions change answers.
Privacy
Runs locally in your browser tab. Snippets stay on your device.
Frequently asked questions
Why do results vary run to run?
GC, JIT tiering and background load — report medians over many iterations.
How many iterations?
Enough that warmup effects vanish: thousands for micro-ops, dozens for heavy ones.
Mean or median?
Median for typical, p99 for tail-sensitive paths.