Bertrand Delsart, Sun Microsystems

Bertrand Delsart is the technical leader for the Sun Java Real-Time System product. He worked mainly on the real-time memory management extensions and on the efficient implementation of the PIP semantic for synchronization primitives. He is now the main contributor to the new real-time garbage collector technology included in Java RTS.

Before joining Sun in 2003, he had previously worked on JVM related issues since the early beginnings of Java at the OSF Research Institute in Grenoble. He investigated most of the JVM components and became the technical leader for all JVM related activities in the lab, including the implementation of a clean room embedded CLDC JVM based on the JEFF class file format and the STIP library (asynchronous library devised for payment terminals).

Presentation: "Real-time Java for latency critical banking applications"

Track: Banking: Complex high volume/low latency architectures

Time: Wednesday 14:30 - 15:30

Location: Abbey Room

Abstract:

Why are you still writing banking applications in C/C++? It's probably not throughput -- compiler improvements made that a non-issue years ago. The number one reason to not use Java these days is garbage collection pause times -- having your application pause in the middle of an order execution can be bad for business. A 1-millisecond advantage in trading applications can be worth millions of dollars a year.

Now, with JVMs that control sources of unpredictability and provide real-time garbage collection, it is possible to retain the productivity, safety, and ease-of-use of the Java platform while offering sufficient determinism for most financial applications. For applications requiring even greater determinism, JVMs that implement the real-time specification for Java provide memory management and APIs to achieve latency/jitter as low as what C/C++ application would achieve. The two memory management schemes can be used within the same application, offering a spectrum of trade-offs of predictability vs programming effort.

In this session, we will explain how Real-Time Java differs from the core Java platform and show how it can be applied to financial applications requiring a higher degree of determinism and predictability.