JDK7新特性之fork/join框架

The fork/join framework is an implementation of the ExecutorService interface that helps you take advantage of multiple processors. It is designed for work that can be broken into smaller pieces recursively. The goal is to use all the available processing power to enhance the performance of your application.

简单翻一下主题思想：fork/join框架是ExecutorService接口的一个实现，它的目的是解决线程的派生问题。这个框架的本质是将一个任务分解成多个子任务，每个子任务用单独的线程去处理。这里用到了递归的思想。

As with any ExecutorService implementation, the fork/join framework distributes tasks to worker threads in a thread pool. The fork/join framework is distinct because it uses a work-stealing algorithm. Worker threads that run out of things to do can steal tasks from other threads that are still busy.

The center of the fork/join framework is the ForkJoinPool class, an extension of the AbstractExecutorService class. ForkJoinPool implements the core work-stealing algorithm and can execute ForkJoinTask processes.

Basic Use

The first step for using the fork/join framework is to write code that performs a segment of the work. Your code should look similar to the following pseudocode:

if (my portion of the work is small enough)
  do the work directly
else
  split my work into two pieces
  invoke the two pieces and wait for the results

Wrap this code in a ForkJoinTask subclass, typically using one of its more specialized types, either RecursiveTask (which can return a result) or RecursiveAction.

After your ForkJoinTask subclass is ready, create the object that represents all the work to be done and pass it to the invoke() method of a ForkJoinPool instance.

Blurring for Clarity

To help you understand how the fork/join framework works, consider the following example. Suppose that you want to blur an image. The original source image is represented by an array of integers, where each integer contains the color values for a single pixel. The blurred destination image is also represented by an integer array with the same size as the source.

Performing the blur is accomplished by working through the source array one pixel at a time. Each pixel is averaged with its surrounding pixels (the red, green, and blue components are averaged), and the result is placed in the destination array. Since an image is a large array, this process can take a long time. You can take advantage of concurrent processing on multiprocessor systems by implementing the algorithm using the fork/join framework. Here is one possible implementation:

public class ForkBlur extends RecursiveAction {
    private int[] mSource;
    private int mStart;
    private int mLength;
    private int[] mDestination;
  
    // Processing window size; should be odd.
    private int mBlurWidth = 15;
  
    public ForkBlur(int[] src, int start, int length, int[] dst) {
        mSource = src;
        mStart = start;
        mLength = length;
        mDestination = dst;
    }

    protected void computeDirectly() {
        int sidePixels = (mBlurWidth - 1) / 2;
        for (int index = mStart; index < mStart + mLength; index++) {
            // Calculate average.
            float rt = 0, gt = 0, bt = 0;
            for (int mi = -sidePixels; mi <= sidePixels; mi++) {
                int mindex = Math.min(Math.max(mi + index, 0),
                                    mSource.length - 1);
                int pixel = mSource[mindex];
                rt += (float)((pixel & 0x00ff0000) >> 16)
                      / mBlurWidth;
                gt += (float)((pixel & 0x0000ff00) >>  8)
                      / mBlurWidth;
                bt += (float)((pixel & 0x000000ff) >>  0)
                      / mBlurWidth;
            }
          
            // Reassemble destination pixel.
            int dpixel = (0xff000000     ) |
                   (((int)rt) << 16) |
                   (((int)gt) <<  8) |
                   (((int)bt) <<  0);
            mDestination[index] = dpixel;
        }
    }
  
  ...

Now you implement the abstract compute() method, which either performs the blur directly or splits it into two smaller tasks. A simple array length threshold helps determine whether the work is performed or split.

protected static int sThreshold = 100000;

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));
}

If the previous methods are in a subclass of the RecursiveAction class, then setting up the task to run in a ForkJoinPool is straightforward, and involves the following steps:

1. Create a task that represents all of the work to be done.

   // source image pixels are in src
   // destination image pixels are in dst
   ForkBlur fb = new ForkBlur(src, 0, src.length, dst);

2. Create the ForkJoinPool that will run the task.

   ForkJoinPool pool = new ForkJoinPool();

3. Run the task.

   pool.invoke(fb);

For the full source code, including some extra code that creates the destination image file, see the ForkBlur example.

Standard Implementations

Besides using the fork/join framework to implement custom algorithms for tasks to be performed concurrently on a multiprocessor system (such as the ForkBlur.java example in the previous section), there are some generally useful features in Java SE which are already implemented using the fork/join framework. One such implementation, introduced in Java SE 8, is used by the java.util.Arrays class for its parallelSort() methods. These methods are similar to sort(), but leverage concurrency via the fork/join framework. Parallel sorting of large arrays is faster than sequential sorting when run on multiprocessor systems. However, how exactly the fork/join framework is leveraged by these methods is outside the scope of the Java Tutorials. For this information, see the Java API documentation.

Another implementation of the fork/join framework is used by methods in the java.util.streams package, which is part of Project Lambda scheduled for the Java SE 8 release. For more information, see the Lambda Expressions section.