Java 7: Meet the Fork/Join Framework | Java Code Geeks

The Fork/Join framework focuses on using all the processing resources available in the machine to improve the performance of the applications. It was designed to simplify parallelism in Divide and Conquer algorithms. The magic behind the Fork/Join framework is its work-stealing algorithm in which work threads that are free steal tasks from other busy threads, so all threads are working at all times. Following are the basics you should know in order to start using the framework:

• Fork means splitting the task into subtasks and work on them.
• Join means merging the solution of every subtask into one general solution.
• java.lang.Runtime use this class in order to obtain the number of processors available to the Java virtual machine. Use the method +availableProcessors():int in order to do so.
• java.util.concurrent.ForkJoinPool Main class of the framework, is the one that implements the work-stealing algorithm and is responsible for running the tasks.
• java.util.concurrent.ForkJoinTask Abstract class for the tasks that run in a java.util.concurrent.ForkJoinPool. Understand a task as a portion of the whole work, for example, if you need to to do something on an array, one task can work on positions 0 to n/2and another task can work on positions (n/2) +1 to n-1, where n is the length of the array.
• java.util.concurrent.RecursiveAction Subclass of the abstract task class, use it when you don’t need the task to return a result, for example, when the task works on positions of an array, it doesn’t return anything because it worked on the array. The method you should implement in order to do the job is compute():void, notice the void return.
• java.util.concurrent.RecursiveTask Subclass of the abstract task class, use it when your tasks return a result. For example, when computing Fibonacci numbers, each task must return the number it computed in order to join them and obtain the general solution. The method you should implement in order to do the job is compute():V, where V is the type of return; for the Fibonacci example, V may be java.lang.Integer.

When using the framework, you should define a flag that indicates whether it is necessary to fork/join the tasks or whether you should compute the work directly. 

    protected void compute() {
        if (mLength < sThreshold) {
            computeDirectly();
            return;
        }
        int split = mLength / 2;
        invokeAll(new ForkBlur(mSource, mStart, split, mDestination),
                new ForkBlur(mSource, mStart + split, mLength - split,
                mDestination));
    }
Read full article from Java 7: Meet the Fork/Join Framework | Java Code Geeks