It's rare to see a well-crafted solution to a thorny problem. With multicore hardware becoming the norm a few years ago, we needed a way to take advantage of all those cores. In Java, we already have threads and synchronization constructs, right? What could be easier than throwing together a bunch of interfaces, classes and threads at this problem?

The answer is not whether it is easier or harder. The question should really be about the right way to do it. The realization of Fork/Join implementation in Java is quite an impressive piece of engineering. Engineering as in recognizing the essence of the problem and solving it just right. Of course, the credit also goes to the original work done in Cilk project at MIT.

I won't try to distill the contents of relevant papers in this post. You should really read Doug Lea's paper and the Cilk-5 paper would be a bonus read after that. Link to these papers are right here.

A Java Fork/Join Framework

The Implementation of the Cilk-5 Multithreaded Language

The design of work-stealing and the implementation of underlying deque is full of deep insights into the problem and clever solutions. I never thought Dekker's algorithm, which provides mutual exclusion between two processes, would ever be usable in a real-world solution. I suppose, some solutions are waiting for just the right problem.