Building a financial exchange like the one that LMAX Exchange operates presents a unique challenge. Our customers trade over $2 trillion a year through our platform, while demanding low and predictable latency with ever increasing volumes of transactions. In this talk we will present an overview of the architecture we use to deliver this, and share our experience of how taking a scientific approach to testing and tuning software has helped us to build a high-availability stateful system that regularly processes many tens of thousands of transactions per second in a single thread, with round-trip latencies through our systems measuring just 80 microseconds - all the while running on commodity Intel servers and written in plain old Java.