ByteBlitz is a CLI tool I built to scratch an itch: I wanted a simple way to benchmark small code snippets without setting up a full testing framework.
What it does
You give it a snippet, it runs it N times, and gives you statistics: mean, median, p95, p99, standard deviation. It handles warmup runs automatically and can compare multiple implementations side by side.
$ byteblitz run --lang python "sorted([3,1,4,1,5,9,2,6])" --iterations 10000
mean: 1.23µs
median: 1.18µs
p95: 1.89µs
p99: 2.34µs
stddev: 0.42µs
runs: 10,000Why I built it
Existing benchmarking tools are either too heavy (full framework setup) or too light (just time in a shell). I wanted something in between — quick to use but statistically rigorous.
Technical details
- Written in Go for fast startup time
- Uses sub-process isolation to prevent benchmark pollution
- Statistical outlier detection using modified Z-scores
- Supports Python, JavaScript, Go, and Rust snippets
- Output formats: table, JSON, CSV
The hardest part was getting accurate timing for sub-microsecond operations. Go’s time.Now() has ~100ns resolution on most systems, so for very fast snippets you need to batch iterations and divide.