在一个约定点同步任务（Synchronizing tasks in a  common point)

Java 并发API中提供了一个同步工具CyclicBarrier类可以使多个线程在约定点进行任务同步，该类和CountDownLatch类有点类似，CountDownLatch是等待多个并发事件，在上节有解释；

CyclicBarrier类初始化有两个参数，第一个是要同步的线程个数，第二个要同步的任务（该参数实现了Runnable接口)；当这些要同步的线程到达到了这个约定的同步点，它将调用await()方法进入睡眠状态，当所有要同步的线程都到达了这个约定的点后，CyclicBarrier将会唤醒所有睡眠的这些线程，然后执行要同步的任务（传入的第二个参数）；

CyclicBarrier最好的就是它可以传入一个实现了Runnable接口的对象，在所有线程到达约定点后，在执行这个对象；这个特性非常适合分治算法 的思想；

再接下来的例子中，简单的使用CyclicBarrier模拟一个简单的分治算法；

问题：即模拟一个矩阵，从该矩阵中查找一个给定的数，并统计该数值在该矩阵中出现的次数；

解决思路：

开启多个线程，每个线程负责查找若干行，并把每一行该数值出现的次数放置到一个数组中，作为结果，每行对应与该数组的索引；在所有线程都查找完成后，执行同步任务，这里仅仅简单的输出最终的结果；

动手实现

1.辅助类，用来模拟矩阵

public class MatrixMock {    private int data[][];    public MatrixMock(int rows,int cols,int number){        int counter=0;        data=new int[rows][cols];        Random random=new Random();        for (int i = 0; i < rows; i++) {            for(int j=0;j
    
     = 0) && (row < data.length)) {            return data[row];        }        return null;    }}
    

2.辅助类，用来放置每个线程处理结果，索引对应矩阵行数，列对应每行查找到指定数值的次数

public class Result {    private int data[];    public Result(int size) {        this.data = new int[size];    }    public void setData(int position,int value){        data[position]=value;    }    public int[] getData(){        return data;    }}

3.用来执行计算的线程

public class Searcher implements Runnable {    private int firstRow;    private int lastRow;    private MatrixMock mock;    private Result result;    private int number;    private final CyclicBarrier barrier;    public Searcher(int firstRow, int lastRow, MatrixMock mock, Result result,                    int number, CyclicBarrier barrier) {        this.firstRow = firstRow;        this.lastRow = lastRow;        this.mock = mock;        this.result = result;        this.number = number;        this.barrier = barrier;    }    @Override    public void run() {        int counter;        System.out.printf("%s: Processing lines from %d to %d.\n",                Thread.currentThread().getName(), firstRow, lastRow);        for (int i = firstRow; i < lastRow; i++) {            int row[] = mock.getRow(i);            counter = 0;            for (int j = 0; j < row.length; j++) {                if (row[j] == number) {                    counter++;                }            }            result.setData(i, counter);        }        System.out.printf("%s: Lines processed.\n", Thread.currentThread().getName());        try {            barrier.await();        } catch (BrokenBarrierException | InterruptedException e) {            e.printStackTrace();        }    }}

4.同步任务，当所有线程完成之后执行，类似于分治算法中的 结果合并；

public class Grouper implements Runnable {    private Result result;    public Grouper(Result result) {        this.result = result;    }    @Override    public void run() {        int finalResult=0;        System.out.printf("Grouper: Processing results...\n");        int data[]=result.getData();        for(int number:data){            finalResult+=number;        }        System.out.printf("Grouper: Total result: %d.\n",finalResult);    }}

5.Main

public class Main {    public static void main(String[] args) {        final int rows=10000;        final int cols=1000;        final int search=5;        final int participants=5;        final int linesParticipant=2000;        MatrixMock mock=new MatrixMock(rows, cols,search);        Result result=new Result(rows);        Grouper grouper=new Grouper(result);        CyclicBarrier barrier=new CyclicBarrier(participants,grouper);        Searcher searchers[]=new Searcher[participants];        for (int i=0; i

一次运行结果：

Mock: There are 1001252 occurrence of 5 in generated data.

Main: The main thread has finished. Thread-0: Processing lines from 0 to 2000. Thread-4: Processing lines from 8000 to 10000. Thread-1: Processing lines from 2000 to 4000. Thread-3: Processing lines from 6000 to 8000. Thread-2: Processing lines from 4000 to 6000. Thread-1: Lines processed. Thread-3: Lines processed. Thread-2: Lines processed. Thread-0: Lines processed. Thread-4: Lines processed. Grouper: Processing results... Grouper: Total result: 1001252.

要点

1.CyclicBarrier还提供了getNumberWaiting()方法，用来获取当前被阻塞的线程个数；

2.利用该类执行分治任务是一个很不错的选择；