Keeping up with the market

As markets grow, our trading infrastructure needs to process ever growing amounts of data in ever shorter time windows. That’s why we build highly-optimized packet processing systems that are capable of handling millions of multicast messages per second on a single core. Building this kind of system requires a disciplined approach to measurement, a focus on determinism and tail-events, and a good dose of mechanical sympathy.

Performance isn’t just important for the most latency-sensitive trading. We’ve built a distributed systems framework based on state machine replication (and inspired by the architecture of financial exchanges) which provides high throughput, low latency, and strong reliability guarantees to a wide variety of internal applications. The architecture of this system depends on a very high-performance backbone for sequencing, distributing, and filtering the transactions that drive these applications.

At the lowest level, we use FPGA accelerators as a way of achieving performance that can’t be gotten on CPUs alone. We lead development of Hardcaml arxiv.org Hardcaml: an ocaml hardware domain-specific language for efficient and robust design , an open-source hardware design library github.com Hardcaml is an OCaml library for designing hardware. - janestreet/hardcaml . By building our own tools, we’ve been able to build a highly productive hardware design workflow, with fast feedback for engineers, and integrated simulation and testing. If you’re interested in seeing Hardcaml at work, some of our engineers won a competition about accelerating zero-knowledge cryptography using it, and we’ve posted the detailed results zprize.hardcaml.com In 2022, we, the team who develops Hardcaml, participated in the ZPrize competition. We comp... .