🏎️ Microbenchmark Framework

A lightweight OCaml benchmarking framework with automatic calibration, statistical analysis, and comparison charts.

Features

Auto-calibration — determines iteration count for reliable measurements
Warmup runs — stabilizes CPU caches before measuring
Full statistics — mean, median, std deviation, percentiles, min/max
Group comparisons — race multiple implementations head-to-head
Parametric benchmarks — measure scaling behavior across input sizes
ASCII visualization — bar charts right in the terminal

Interactive Demo

Simulated benchmark results (the real framework runs natively in OCaml):

API Usage

(* Simple benchmark *)
let result = Bench.run "My Function" (fun () -> my_function input)

(* Compare implementations *)
Bench.group "Sorting" [
  ("Insertion Sort", fun () -> insertion_sort data);
  ("Merge Sort",     fun () -> merge_sort data);
  ("Array.sort",     fun () -> Array.sort compare (Array.copy data));
]

(* Parametric — measure scaling *)
Bench.parametric "Array.sort Scaling"
  [100; 500; 1000; 5000; 10000]
  (fun n ->
    let data = Array.init n (fun _ -> Random.int 100000) in
    fun () -> Array.sort compare (Array.copy data))

(* Custom config *)
let cfg = { Bench.default_config with
  warmup_iters = 10;
  min_samples = 50;
  min_time = 3.0;
} in
Bench.run ~config:cfg "Precise" my_fn

Statistics Explained

x̄

Mean — average time

x̃

Median — middle value

Std Dev — spread

5th percentile

P95

95th percentile

Sample count

← Back to Index