Softonaut.com

JavaPolis Day Three

December 14th, 2006 Peter

JavaPolis University has closed its doors, giving way to the main conference. The latter is more about breadth than depth, with many short (typically one-hour) sessions instead of a few extensive ones. Among the more interesting sessions today was Guy Nirpaz from GigaSpaces talking about SOA in a stateful low-latency environment. Before presenting their own space-based approach, he discussed other ways of building SOAs, using technologies such as JMS, Jini, ESBs, and of course Web Services.
Today’s brilliancy price goes to Brian Goetz for his talks on “Java Concurrency” and “Java Performance Myths”. I’d like to elaborate on the latter, because it very much complies with my own beliefs. Brian showed lots of nice examples demonstrating that in the presence of modern JVMs, low-level performance tricks that we are used to from C and often also employ in Java programs almost never pay off. Or to put it in Brian’s words: “Don’t try to outsmart the JVM, or you will lose!”. Why? Because JVMs already go to incredibly great lengths to optimize your code, and your own efforts might interfer with theirs. So unless you know exactly what you do (which basically means you are on Sun’s JVM team), leave the low-level stuff to them! Otherwise you might end up with worse rather than better performance, while at the same time complicating if not breaking your code (remember the double-checked locking idiom?). Instead, spend your time on practicing a clean and simple programming style, and pay attention to good high-level design. This will ultimately take you much further, even in terms of performance. Thanks for the pleasure, Brian!
In the evening, Mario and I attended the “Meet your Java Idol” event. While not many of them showed up, at least there was free beer for everyone.

Entry Filed under: JavaPolis06