Persistence apparently pays off. The P99 CONF community has been requesting Daniel Lemire for a few years now, and the stars finally aligned. In just about a month (!), Daniel and Francisco Geiman Thiesen will be presenting Billion-Query-Per-Second Lookup Tables in C++: Compile-Time Perfect Hashing.
Note: P99 CONF is a technical conference on performance and low-latency engineering. It’s virtual, free, and highly interactive. This year’s agenda spans Rust, Tokio, Zig, C++, Go, eBPF, observability, io_uring, kernel, distributed systems, and of course AI/ML.
We hope you’ll join us live on October 21-22 to hear the talk and chat with Daniel (and Francisco). In the meantime, let’s get to know more about him!
How do you answer the dreaded “tell us about yourself” question?
I am a weird person. I am both a software developer and a university professor, which is unusual by itself. I have been blogging for over 20 years, a rare feat. I failed kindergarten, I did not learn my multiplication tables, but I got a PhD in math. I have spent the last 20 years working on a topic that is almost forgotten on campus: software performance.
What’s the most interesting project that you’re working on right now – or hoping to start soon?
I hope to release soon, a version of our fast unicode library (simdutf) that will support accelerated transcoding with replacement. This library is found in mainstream browsers and our new code should help simplify and accelerate critical code.
What will you be talking about at P99 CONF?
How to do string lookups in tables in C++ at crazy speeds.
Editor’s note: Here’s their full abstract:
In high-performance systems, string-to-value lookups (HTTP methods, headers, protocols, keywords, configuration keys, stock tickers, etc.) often sit on the hot path. Even “fast” hash tables add measurable overhead under load, especially with variable-length keys.
When the set of keys changes rarely, we can use “perfect” hashing techniques. With these things, we can minimize the memory usage and avoid the collisions that occur when two keys have the same hash value.
This talk presents a modern C++23 single-header library that builds high performance perfect hash functions entirely at compile time using immediate functions. We designed it for speed. The result? We achieve over 1 billion queries per second on realistic key sets (URL protocols, HTTP headers, etc.): 4× faster than the closest competitor and 10× faster than std::unordered_map. The library supports both sets and maps, with a clean modern C++ APIs.
We cover:
- Practical usage examples (parsing HTTP methods, headers, protocols, etc.)
- How to integrate into large codebases with precompiled headers (PCH) for negligible compile-time impact
- Benchmark design and results against gperf, frozen, absl::flat_hash_map, ankerl::dense_hash_map, pthash, and more, across small-to-medium key sets with hits, misses, and mixed workloads
- Trade-offs, limitations, and when this approach crushes traditional hash tables (or gperf)
- Data parallel (SIMD) optimizations
- Applications to C++26 compile-time reflection
Our technique eliminates a common source of tail latency in parsers, routers, deserializers, market-data pipelines, and hot-path state machines where inputs map to known symbol sets (such as protocol names, headers, or tickers). If you’re obsessed with shaving nanoseconds and building predictable low-latency systems in C++, this talk delivers concrete, immediately usable techniques and code you can adopt today. Our code will be open source.
What other P99 CONF talks are you most looking forward to – and why?
There are many, but my friend Tomás Senart from Perfloop is surely going to be a must watch !
What do you like most about P99 CONF?
I think that it is the best conference right on software performance, taken broadly.
Any performance-related blog/article/paper/book/video recommendations for the P99 CONF community?
Please buy my latest book: Mastering Programming: From Testing to Performance in Go
https://www.amazon.com/dp/B0FMPGSWR5/
I got a lot of positive feedback, people like it.
You also must subscribe to my blog at http://lemire.me/blog/
