disty

disty is a quick CLI for getting an idea of the distribution of a list of numbers.

disty is a rewrite of Michael Knyszek's distx, which is itself an extension of Austin Clements' dist. I use distx for all sorts of quick numeric checks, from checking the distribution of database segments to analyzing request latency. I ran into performance issues for processing large lists (>1m records) which inspired the rewrite.

Compared to distx, this version:

• Is very fast: ~55x faster in local testing. This boils down to parallelizing the KDE plotting, using mmap to parallelize parsing the number list, and reducing unnecessary copying.

  $ seq 1 10000000 | rg 1 > /tmp/large_seq
  $ wc -l /tmp/large_seq
  5217032 /tmp/large_seq
  $ hyperfine --warmup 3 'distx /tmp/large_seq' 'disty /tmp/large_seq'
  Benchmark 1: distx /tmp/large_seq
    Time (mean ± σ):     10.750 s ±  0.625 s    [User: 10.711 s, System: 0.471 s]
    Range (min … max):    9.546 s … 11.438 s    10 runs
  
  Benchmark 2: disty /tmp/large_seq
    Time (mean ± σ):     195.6 ms ±   1.9 ms    [User: 1757.8 ms, System: 153.9 ms]
    Range (min … max):   191.3 ms … 199.4 ms    15 runs
  
  Summary
    disty /tmp/large_seq ran
     54.95 ± 3.24 times faster than distx /tmp/large_seq    
  

• Has marginally better plotting, which mostly comes down to setting a higher resolution than distx uses by default.

• Has less features. I haven't ported the output options or alternative plotting (CDFs), because I don't really use them.

Installing

Install via Cargo.

# From Crates.io
$ cargo install disty-cli
# From source
$ cargo install --path .

Usage

$ disty --help
Summarizes numerical distributions

Usage: disty [OPTIONS] [INPUT]

Arguments:
  [INPUT]  Input file (stdin if not specified)

Options:
  -u, --unit <UNIT>  Input unit [possible values: ns, us, ms, s, B, KB, MB, GB, TB, PB, KiB, MiB, GiB, TiB, PiB]
  -f, --fmt <FMT>    Output format [possible values: float, hex, time, bytes]
      --no-plot      Skip KDE plotting
  -h, --help         Print help
  -V, --version      Print version

Development

Running Tests

cargo test

Running Benchmarks

The project includes criterion benchmarks for parsing, statistics computation, and KDE evaluation:

# Run all benchmarks
cargo bench

# Run specific benchmark suite
cargo bench --bench disty_bench -- parsing
cargo bench --bench disty_bench -- stats
cargo bench --bench disty_bench -- kde
cargo bench --bench disty_bench -- full_pipeline

# Run benchmarks with different sample sizes
cargo bench --bench disty_bench -- "1000000"

Benchmarks test with various input sizes (1K, 10K, 100K, 1M elements) to understand performance characteristics at different scales.