Forking and Joining Java to Maximize Multicore Power

f writing software is hard, then writing applications that will take full advantage of the ever-more-prevalent multicore/multi-CPU hardware—without introducing subtle race conditions and consistency errors—is even harder. Numerous solutions have been floated for Java as well as other environments, solutions such as software transactional memory API libraries, functional languages, and even entirely new environments. However, no solution has emerged as the clear winner or even clear front-runner yet.

For its part, Sun has offered language features for structuring concurrency applications. Java 5 introduced a slew of concurrency data structures as well as behaviors and locking tools as part of the java.util.concurrent package, and Java 6 added a few more thread-safe data structures. Now Sun is set to release the first "new" approach to concurrency since the release of the Thread API in Java 1.0. Based on the work Doug Lea began back in 2000 to port a "work-stealing" mechanism to Java, this new API is known officially as JSR-166y (to distinguish it from the original JSR-166 for Java 5). To the initiated, it is colloquially called the fork/join framework.