相位和CyclicBarrier的区别

Question

我无意中发现了CyclicBarrier和Phaser实用程序在中的区别。

我理解CyclicBarrier允许一组线程等待直到所有线程到达一个特定的点。相位器也这样做，但它支持多个阶段。我也明白，CyclicBarrier可以重用。--我认为这个重用工具使它的功能与Phaser相同。

考虑下列方案：

测试相位:

import java.util.concurrent.Phaser;

public class PhaserTest {

    public static void main(String[] args) {
        Phaser p = new Phaser(3);
        Thread t1 = new Thread(() -> process(p), "T1");
        Thread t2 = new Thread(() -> process(p), "T2");
        Thread t3 = new Thread(() -> process(p), "T3");
        t1.start();
        t2.start();
        t3.start();
    }

    private static void process(Phaser p) {
        try {
            System.out.println("Started Phase 1: "+Thread.currentThread().getName());
            p.arriveAndAwaitAdvance();
            System.out.println("Finished Phase 1: "+Thread.currentThread().getName());
            System.out.println("Started Phase 2: "+Thread.currentThread().getName());
            p.arriveAndAwaitAdvance();
            System.out.println("Finished Phase 2: "+Thread.currentThread().getName());
        } catch(Exception e) {}
    }

}

输出：

Started Phase 1: T1
Started Phase 1: T2
Started Phase 1: T3
Finished Phase 1: T3
Started Phase 2: T3
Finished Phase 1: T1
Finished Phase 1: T2
Started Phase 2: T2
Started Phase 2: T1
Finished Phase 2: T2
Finished Phase 2: T3
Finished Phase 2: T1

测试CyclicBarrier:

import java.util.concurrent.CyclicBarrier;


public class CyclicBarrierTest {
    public static void main(String[] args) {    
        CyclicBarrier cb = new CyclicBarrier(3);
        Thread t1 = new Thread(() -> process(cb), "T1");
        Thread t2 = new Thread(() -> process(cb), "T2");
        Thread t3 = new Thread(() -> process(cb), "T3");
        t1.start();
        t2.start();
        t3.start();
    }

    private static void process(CyclicBarrier cb) {
        try {
            System.out.println("Started Phase 1: "+Thread.currentThread().getName());
            cb.await();
            System.out.println("Finished Phase 1: "+Thread.currentThread().getName());
            System.out.println("Started Phase 2: "+Thread.currentThread().getName());
            cb.await();
            System.out.println("Finished Phase 2: "+Thread.currentThread().getName());
        } catch(Exception e) {}
    }
}

输出：

Started Phase 1: T1
Started Phase 1: T2
Started Phase 1: T3
Finished Phase 1: T3
Started Phase 2: T3
Finished Phase 1: T1
Started Phase 2: T1
Finished Phase 1: T2
Started Phase 2: T2
Finished Phase 2: T2
Finished Phase 2: T3
Finished Phase 2: T1

在PhaserTest和CyclicBarrierTest中，直到所有各方都到达/完成前一阶段，下一阶段才开始。

所以，拥有Phaser有什么好处?

Answer 1

相位器来自JSR-166

在将阶段器与现有Java特性进行比较时，我们解释了支持与CyclicBarrier类类似的功能(在Java 5中引入)，但是phasers本质上更灵活： java.util.concurrent CyclicBarrier类支持一组线程之间的周期性屏障同步。然而，与相位器不同的是，CyclicBarriers不支持线程的动态添加或删除，也不支持单向同步或分阶段操作。探索其他障碍实现的主要动机之一不仅是增强灵活性，还在于提高屏障同步概念的性能和可伸缩性： 在三个不同的SMP平台上实现阶段器的可移植实现所获得的性能结果表明，除了它们的通用性和安全性属性带来的生产力好处之外，它们还可以提供比现有障碍实现更好的性能。

这里的全部解释：https://www.infoq.com/news/2008/07/phasers

还有一篇将Phaser定义为CyclicBarier+CountdownLatch的比较文章：http://flex4java.blogspot.com/2015/03/why-and-how-to-use-phaser-in-java.html