Martin has had a passion for pushing software and electronics to the limit since childhood. He was the type of kid who took the video recorder apart and then fixed it. Since then he has always been attracted to business problems where high performance computing can open new, previously impossible, opportunities. Ranging from stock market data feeds to PCs as in the early 90s, working on the first generation of internet banks, implementing the largest product content management systems, working on the worlds largest sports betting exchange (Betfair) and now founding LMAX/Tradefair. Martin brings his mechanical sympathy for the hardware that it runs on, to the software that he creates, which has taken him deep into the subjects of concurrency and parallel computing. Martin is the co-founder and CTO of LMAX where he leads the building of the world’s highest performance financial exchange.
Presentation: "LMAX - How to do over 100K contended complex business transactions per second at less than 1ms latency"
Time: Thursday 15:00 - 16:00
Location: St. James's Suite
Conventional wisdom says that memory is fast, disk is slow, networks are slower and that fast systems must be highly concurrent to achieve maximum performance. Much of this is outdated and some of it is now wrong. Modern hardware is phenomenally fast, but we have become complacent and use it in extremely inefficient and inappropriate ways.
For world class performance on commodity hardware you need to take a holistic approach to software design. The good news is that all the stuff we learned in computer science 101 is what really matters, choose your data structures carefully, model your domain, understand your platform, work hard to have a clear separation of concerns, but as well as all of that, run your business logic on a single thread!?
Martin Thompson and Dave Farley work for the London Multi-Asset Exchange and are building one of the world's fastest trading platforms using a radical new architecture that takes it all back to basics. This session will describe what you need to do to develop the worlds fastest financial exchange - little things like how to do over 100K/tps at less than 1ms latency and still have full HA support in a distributed environment.
Keywords: Java, Architecture, Low Latency, Performance, Exchange, Concurrency Patterns
Target Audience: Anyone interested in high performance low-latency architecture or concurrency