Parsing 1B rows in Ocaml

Inspired by the 1 Billion Row Challenge in Java and 1brc by taekim in Bun/TypeScript, I wanted to implement it in OCaml (and try some of the OxCaml features).

Current best run (results/2026-03-14_20-23-25_136617877.txt):

• Total rows processed: 1,000,000,000
• Total time (wall clock): 3.49s
• Internal timer (Processed in): 0.87s
• Average throughput (wall clock): 286,532,951 rows/second

System specs

• OS: Linux 6.12.22+bpo-amd64 (x86_64)
• CPU: 2x AMD EPYC 9755 128-Core Processor (256 cores / 512 threads)
• Memory: 2.2TiB RAM
• OCaml: 5.2.0+ox
• Dune: 3.21.1

Instructions

Step 0: Use dune pkg manager

Install dune 3.21.1

Step 1: Create the file with 1 billion rows

Install dependencies, this project is configured to work with OxCaml (OCaml 5.2.0+ox).

make install

Generate the measurements file:

make generate

NOTE: file is about 13.8GB (data/measurements.txt), make sure you have enough space.

Step 2: Run the parser

Build the project:

make build

Run it on a test file first:

dune exec ./main.exe -- data/small.txt

Run it on measurements.txt (the 13.8GB file):

/usr/bin/time -v dune exec ./main.exe -- data/measurements.txt