Nitin Verma’s Blog

Using jdk 1.6 the cost of blocking synchronization is nearly same as
non-blocking counter part. This does not say 1.6 makes it liveness
hazard free or deadlock free.

But makes it quite fast.

// We have three methods
// -1: Thread unsafe
    public static void addUnsafe() {
        b++;
    }
// 0: Thread safe, non-blocking synchronization
    public static void addNBSync() {
        try {
            lock.lock();
            b++;
        } finally {
            lock.unlock();
        }
    }
// 1: Thread safe, blocking synchronization
    public static synchronized void addSync() {
        b++;
    }

// Log format
//Use:<method>:<end value of b>:<time taken>
$ ~/usr/local/jdk1.5.0_17/bin/java -classpath classes1.5  edu.nitin.Main
Use:-1:63929500:156367000
Use:0:100000000:5360473000
Use:1:100000000:24290461000

$ echo "scale=10; ( 5360473000 - 156367000 ) / 156367000 " | bc
33.2813573196
$ echo "scale=10; ( 24290461000 - 156367000 ) / 156367000 " | bc
154.3426298387

$ ~/usr/local/jdk1.6.0_11/bin/java -classpath classes1.6  edu.nitin.Main
Use:-1:59050465:190538454
Use:0:100000000:5350310972
Use:1:100000000:6404364288

$ echo "scale=10; ( 5350310972 - 190538454 ) / 156367000 " | bc
32.9978353361
$ echo "scale=10; ( 6404364288 - 190538454 ) / 156367000 " | bc
39.7387289773


A test running 1000 threads doing 100,000 atomic operations each,
all the three ways.
...
package edu.nitin;
import java.util.concurrent.locks.Lock;

import java.util.concurrent.locks.ReentrantLock;
/**

 *

 * @author Nitin Verma

 */

public class Main {
    private final static Lock lock = new ReentrantLock();
    private static long b = 0;

    private final static long n = 100000;

    private final static int a = 1000;
    /**

     * @param args the command line arguments

     */

    public static void main(String[] args) throws InterruptedException {

        test(-1);

        test(0);

        test(1);

    }
    private static void test(final int use) throws InterruptedException {

        b = 0;

        final long start = System.nanoTime();

        long end = 0;

        try {

            final Thread[] ta = new Thread[a];

            for (int i = 0; i < a; i++) {
                ta[i] = new Thread(new MyRunnable(use));
                ta[i].start();
            }

            for (int i = 0; i < a; i++) {
                if (ta[i] != null && ta[i].isAlive()) {
                    ta[i].join();
                }
            }
        } finally {
            end = System.nanoTime();
        }
        System.out.println("Use:" + use + ":" + b + ":" + (end - start));
    }

    public static void addN(final long n, final int use) {
        if (use == 0) {
            for (long i = 0; i < n; i++) {
                addNBSync();
            }
        }
        else if (use == 1)  {
            for (long i = 0; i < n; i++) {
                addSync();
            }
        }
        else {
            for (long i = 0; i < n; i++) {
                addUnsafe();
            }
        }
    }

    public static void addNBSync() {
        try {
            lock.lock();
            b++;
        } finally {
            lock.unlock();
        }
    }

    public static synchronized void addSync() {
        b++;
    }

    public static void addUnsafe() {
        b++;
    }

    private static class MyRunnable implements Runnable {

        private int use;

        public MyRunnable(final int use) {
            this.use = use;
        }

        public void run() {
            addN(n, use);
        }
    }
}
[/sourcecode]