Building a high performance financial exchange is tough, especially when you're flying in the face on conventional wisdom. In this case it is widely assumed that it's not possible to build a high-throughput near real-time system in Java. Over the past 4 years our experience of building LMAX, the world's first multi-asset financial exchange, we've learnt that much of what the industry believes about computer performance is based on folklore. Using evidence and experience from solving real world problems, this talk aims to dispel some these performance myths. We will cover a number of areas of software development including distributed systems, concurrency and parallelism, I/O, algorithms, garbage collection, and memory efficiency. We need to reintroduce the science into computer science and stop depending on mythology, technology is changing and mythology is not keeping up.