anytimetraveler / 1brc

The One Billion Row Challenge

This is my implementation for The One Billion Row Challenge.

The goal is, to process 1 billion lines of CSV as fast as possible.

Log

27.04.2024

After an afternoon/evening of casual tinkering and swapping out some map implementations for better ones, I am now at 3.926 seconds.

I wish I had recorded my earlier times.

I do remember that replacing the std::collections::HashMap with halfbrown::HashMap roughly halved my time.
But I suspect that there is more performance to be gained from an array-based approach.

I know of several improvements that I could still make:

• do some profiling to see where the bottlenecks lie
• take advantage of bitwise operations to avoid branches
• write my own optimized hash-map replacement
• write my own float-parser
• get away from float-parsing entirely and just use an integer