Java Magazine - July/August 2012

//java architect /

Fork/Join Framework for
Client Java Applications
Java SE 7’s fork/join framework makes a great match for CPU-intensive client-side applications.
JOSH MARINACCI

JAVA IN ACTION

CPUs aren’t getting faster. Nearly ten years ago, we hit 3 GHz and I have yet to own a

4 GHz machine, much less the 43 GHz Moore’s Law promised me. CPUs can do far more than they used to, but they do it by getting wider instead of faster— more CPUs with more cores with more threads.

While this is great for the end user, it introduces new challenges for application developers. I can buy a 12-core desktop from Apple, and even my small laptop has 2 cores now. The world is becoming parallel, so we need new ways to code for parallel machines.

Java has always had good sup-
port for concurrent programming
thanks to java.lang. Thread. The
Thread class has existed since the
first release of J2SE 5.0 introduced
the ExecutorService to ease the
management of groups of threads.
Unfortunately, threads mainly
help with I/O-bound tasks. That’s
fine for server-side jobs, such as
hosting Web apps, but modern
desktop apps need help with CPU-
bound tasks as well.

if you don’t know the scope of the work beforehand.

■ ■ Fork/join provides work-stealing behavior that can better balance the workload across multiple processors, and with less lock contention. Traditional algorithms can max out at about eight processors before the overhead of lock contention outweighs the speed of using more cores. Fork/join can scale to over 100 cores.

■ ■ Fork/join is future-proof and portable. The code you write with it doesn’t assume anything about the underlying hardware. It’s designed to run your code as efficiently as possible across any hardware—both today’s

Every class that
uses the fork/join
framework works
with pretty much
this algorithm: If the
amount of work is
below a threshold,
then do the work;
other wise, split
the work in half and
recurse, waiting
for each half to
complete.

dual-core laptops and single-core phones, as well as the 100-core desktop of the future.

Something Simple To see how the fork/ join framework works, let’s try something simple. Every class that uses the fork/join framework works with pretty much this algorithm (see Listing 1): If the amount of work is below a threshold, then do the work; otherwise, split the work in half and recurse, waiting for each half to complete.

This is the textbook divide-and-conquer method.

To demonstrate how this works in practice, I’ve created a simple example that calculates the minimum number from a very large array of doubles (see Listing 2).