欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

并发集合ConcurrentSkipListMap的性能测试

程序员文章站 2022-06-30 23:02:27
...

 

ConcurrentSkipListMap是Doug Lea在java6中新加入的一个并发集合,下面这个例子主要是测试ConcurrentSkipListMap的插入、读取和并发修改集合元素时的性能特征,代码如下:

 

 

package test.caipiao.log;

import java.io.File;
import java.io.IOException;
import java.nio.charset.Charset;
import java.util.concurrent.ConcurrentNavigableMap;
import java.util.concurrent.ConcurrentSkipListMap;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.ArrayList;
import java.util.Iterator;
import java.util.Map;
import java.util.Set;

import com.caipiao.util.io.FileUtils;

public class ConcurrentSkipListMapTest2 {
    

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

        final ConcurrentNavigableMap<Integer, String> cslm = new ConcurrentSkipListMap<Integer, String>();
        final Counter5 counter = new Counter5();

        // create 100 threads
        ArrayList<MyThread5> threads = new ArrayList<MyThread5>();
        for (int x = 0; x < 100; x++) {
            threads.add(new MyThread5(counter, cslm));
        }

        long start1 = System.currentTimeMillis();
        // start all of the threads
        Iterator i1 = threads.iterator();
        while (i1.hasNext()) {
            MyThread5 mt = (MyThread5) i1.next();
            mt.start();
        }

        // wait for all the threads to finish
        Iterator i2 = threads.iterator();
        while (i2.hasNext()) {
            MyThread5 mt = (MyThread5) i2.next();
            mt.join();
        }

        long end1 = System.currentTimeMillis();
        System.out.println("100个线程 每个线程插入100000共1000万条记录,耗时: " + (end1 - start1) + " 毫秒");
        System.out.println("Count: " + counter.getCount());
        System.out.println("original size  = " + cslm.size());
        System.out.println("---------------------以上是测试插入---------------------------------------");
        
        // create 1000 threads
        ArrayList<MyThread7> threads3 = new ArrayList<MyThread7>();
        for (int x = 0; x < 1000; x++) {
            threads3.add(new MyThread7(counter, cslm));
        }

        long start3 = System.currentTimeMillis();
        // start all of the threads
        Iterator i3 = threads3.iterator();
        while (i3.hasNext()) {
            MyThread7 mt = (MyThread7) i3.next();
            mt.start();
        }

        // wait for all the threads to finish
        Iterator i33 = threads3.iterator();
        while (i33.hasNext()) {
            MyThread7 mt = (MyThread7) i33.next();
            mt.join();
        }

        long end3 = System.currentTimeMillis();
        System.out.println("1000个线程 每个线程插入10000共1000万条记录,耗时: " + (end3 - start3) + " 毫秒");
        System.out.println("---------------------以上是测试读取---------------------------------------");
        
        long start2 = System.currentTimeMillis();
        ArrayList<MyThread6> threads2 = new ArrayList<MyThread6>();
        for (int x = 0; x < 5; x++) {
            threads2.add(new MyThread6(counter, cslm));
        }

        // start all of the threads
        Iterator i12 = threads2.iterator(); //并发修改 访问 ConcurrentSkipListMap
        while (i12.hasNext()) {
            MyThread6 mt = (MyThread6) i12.next();
            mt.start();
        }
        
        Thread t = new Thread() {
            public void run() {
                for (int x = 0; x < 100000; x++) {
                    String s = cslm.get(x);
                    if (x % 20 == 0) {
//                        System.out.println("key ---" + x + " s--- " + s);
                    }
                }
            }
        };
        
        t.start();
        
        
        
     // wait for all the threads to finish
        Iterator i22 = threads2.iterator();
        while (i22.hasNext()) {
            MyThread6 mt = (MyThread6) i22.next();
            mt.join();
        }
        
        t.join();
        
        long end2 = System.currentTimeMillis();
        System.out.println(end2 - start2);
        System.out.println("over");
        System.out.println("new size  = " + cslm.size());
        System.out.println("---------------------以上是测试并发的修改ConcurrentSkipListMap里的元素---------------------------------------");
        
    }
}

// thread that increments the counter 100000 times.
class MyThread5 extends Thread {
    Counter5 counter;
    ConcurrentNavigableMap<Integer, String> cslm;

    MyThread5(Counter5 counter, ConcurrentNavigableMap<Integer, String> cslm) {
        this.counter = counter;
        this.cslm = cslm;
    }

    public void run() {
        for (int x = 0; x < 100000; x++) {
            int i = counter.incrementCount();
            
            cslm.put(i, "");
        }
    }
}

// class that uses AtomicInteger for counter
class Counter5 {

    private AtomicInteger count = new AtomicInteger(0);

    public int incrementCount() {
        return count.incrementAndGet();
    }
    
    public int decrementAndGet() {
        return count.decrementAndGet();
    }

    public int getCount() {
        return count.get();
    }
}

//class that uses to delete part data of the cslm
class MyThread6 extends Thread {
    Counter5 counter;
    ConcurrentNavigableMap<Integer, String> cslm;

    MyThread6(Counter5 counter, ConcurrentNavigableMap<Integer, String> cslm) {
        this.counter = counter;
        this.cslm = cslm;
    }

    public void run() {
        for (int x = 0; x < 100000; x++) {
            cslm.remove(x);
            try {
                if (x % 100 == 0) {
                    this.sleep(1);
                }
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
        }
    }
}


//thread that decrements the counter 100000 times.
class MyThread7 extends Thread {
    Counter5 counter;
    ConcurrentNavigableMap<Integer, String> cslm;
    
    MyThread7(Counter5 counter, ConcurrentNavigableMap<Integer, String> cslm) {
        this.counter = counter;
        this.cslm = cslm;
    }
    
    public void run() {
        for (int x = 0; x < 10000; x++) {
            int i = counter.decrementAndGet();
            cslm.get(i);
        }
    }
}

 

 

CPU配置为4核 intel i3-2100 3.1GHz,内存分配1G,多次运行输出大致如下:

100个线程 每个线程插入100000共1000万条记录,耗时: 4719 毫秒

Count: 10000000

original size  = 10000000

---------------------以上是测试插入---------------------------------------

1000个线程 每个线程插入10000共1000万条记录,耗时: 2684 毫秒

---------------------以上是测试读取---------------------------------------

1025

over

new size  = 9900001

-------------------以上是测试并发的修改、访问ConcurrentSkipListMap里的元素------------------------------------

 

可以看到put操作的并发能达到200万/秒以上,get操作的并发能达到350万/秒以上,

这里的性能损耗有一部分是在AtomicInteger原子的递增和递减上,如果在真实的业务场景中,可以知道key的操作不需要使用AtomicInteger或加锁方式来获取,性能会更高。

通过第三个输出可以知道对ConcurrentSkipListMap访问的同时进行删除不会出现ConcurrentModificationException异常。

 

此例子可以把ConcurrentSkipListMap换成ConcurrentHashMap等其它集合进行测试。

 

 

 如果把threads的线程数创建为10000个而每个线程执行1000次,是可以正常输出的。

把ConcurrentSkipListMap换成ConcurrentHashMap,在把threads的线程数创建为10000个而每个线程执行1000次,在线程数执行不到5000的时候会报java.lang.OutOfMemoryError: unable to create new native thread,但把每个线程的执行次数改为100的时候,可以正常的执行下去。在ConcurrentHashMap的put操作中有tryLock()这个操作会涉及到线程操作,而在ConcurrentSkipListMap的put操作中没有涉及线程操作,具体原因只有仔细的看看源码才能解释。