When either Alex Petrov or Thorsten Ball recommends a P99 CONF speaker who is not yet on our radar, we’re intrigued. When they BOTH independently recommend the same person within a week, we know we have to welcome this person into the P99 CONF community.
Long story short, that’s how we “discovered” Tomás Senart — and why we’re so eager to host his talk on “Closed-Loop Performance Engineering” this year.
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 Tomás. In the meantime, let’s get to know more about him!
How do you answer the dreaded “tell us about yourself” question?
I’ve been obsessed with software performance for more than 15 years and I get most excited by big ideas that go against the grain.
At a bar, we’d talk about technology, progress and the future. I live in Munich with my wife and my two great boys.
I am building Perfloop.
What’s the most interesting project that you’re working on right now – or hoping to start soon?
Perfloop. A system for end to end performance-engineering: finds the slow code, changes it, proves the number moved and nothing regressed. Most changes don’t make it out. Proving is the scarcer half. The system carries the relentless part, the finding and the proving, with the numbers to back it up, while the person stays in charge of what is worth holding and what actually ships. We are doing that work in open source, in public, with contributions so far across projects like ClickHouse, Prometheus, Roaring Bitmaps, Parquet Go, Zoekt and more. I wrote the first fourteen weeks in [Compute in, proof out].
What will you be talking about at P99 CONF?
A field report on that loop. It watches the running system, decides what to investigate, runs the experiment, returns a verdict the team can audit, and learns across cycles. The method is still hypotheses, tests, and hard data. Agents can run the work between those steps, so the experiments go faster. I will say which pieces reach autonomy today and which ones still need a person.
What other P99 CONF talks are you most looking forward to – and why?
Daniel Lemire and Francisco Geiman Thiesen on compile-time perfect hashing. A billion lookups a second from a table the compiler already finished. I want the library, and I want to see how they build it.
Dominik Tornow, The Prompt is the Platform. Spec, verify, optimize, then an agent builds Resonate on ScyllaDB. I want to see how far that loop gets.
Nathan VanBenschoten on searching 100 billion vectors on object storage at 200ms p99. At Axiom I built Haydex, a system for trillion-row search over petabyte event data. I want to see how they hold 200ms p99 when that much data lives on object storage.
What do you like most about P99 CONF?
The intense focus on performance. I sent an abstract because I wanted a room that loves that topic as much as I do.
Any performance-related blog/article/paper/book/video recommendations for the P99 CONF community?
Gil Tene, How NOT to Measure Latency. I still send it to people. Coordinated omission is why Vegeta does not wait on the server before issuing the next request.
The other line I keep repeating: the benchmark you ran once is already stale. Performance is not something you set and keep, it is an envelope you have to hold against constant change. That is in the Heavybit interview if you want the longer version.
