首页 > 学院 > 开发设计 > 正文

各种同步方法性能比较(synchronized,ReentrantLock,Atomic)

2019-11-14 10:39:26
字体:
来源:转载
供稿:网友

转自:http://zzhonghe.VEvb.com/blog/826162

5.0的多线程任务包对于同步的性能方面有了很大的改进,在原有synchronized关键字的基础上,又增加了ReentrantLock,以及各种Atomic类。了解其性能的优劣程度,有助与我们在特定的情形下做出正确的选择。 

总体的结论先摆出来:  

synchronized: 

在资源竞争不是很激烈的情况下,偶尔会有同步的情形下,synchronized是很合适的。原因在于,编译程序通常会尽可能的进行优化synchronize,另外可读性非常好,不管用没用过5.0多线程包的程序员都能理解。 

ReentrantLock: 

ReentrantLock提供了多样化的同步,比如有时间限制的同步,可以被Interrupt的同步(synchronized的同步是不能Interrupt的)等。在资源竞争不激烈的情形下,性能稍微比synchronized差点点。但是当同步非常激烈的时候,synchronized的性能一下子能下降好几十倍。而ReentrantLock确还能维持常态。 

Atomic: 

和上面的类似,不激烈情况下,性能比synchronized略逊,而激烈的时候,也能维持常态。激烈的时候,Atomic的性能会优于ReentrantLock一倍左右。但是其有一个缺点,就是只能同步一个值,一段代码中只能出现一个Atomic的变量,多于一个同步无效。因为他不能在多个Atomic之间同步。 所以,我们写同步的时候,优先考虑synchronized,如果有特殊需要,再进一步优化。ReentrantLock和Atomic如果用的不好,不仅不能提高性能,还可能带来灾难。 先贴测试结果:再贴代码(Atomic测试代码不准确,一个同步中只能有1个Actomic,这里用了2个,但是这里的测试只看速度) ========================== round:100000 thread:5 Sync = 35301694 Lock = 56255753 Atom = 43467535 ========================== round:200000 thread:10 Sync = 110514604 Lock = 204235455 Atom = 170535361 ========================== round:300000 thread:15 Sync = 253123791 Lock = 448577123 Atom = 362797227 ========================== round:400000 thread:20 Sync = 16562148262 Lock = 846454786 Atom = 667947183 ========================== round:500000 thread:25 Sync = 26932301731 Lock = 1273354016 Atom = 982564544 java代码  收藏代码package test.thread;    import static java.lang.System.out;    import java.util.Random;  import java.util.concurrent.BrokenBarrierException;  import java.util.concurrent.CyclicBarrier;  import java.util.concurrent.ExecutorService;  import java.util.concurrent.Executors;  import java.util.concurrent.atomic.AtomicInteger;  import java.util.concurrent.atomic.AtomicLong;  import java.util.concurrent.locks.ReentrantLock;    public class TestSyncMethods {            public static void test(int round,int threadNum,CyclicBarrier cyclicBarrier){          new SyncTest("Sync",round,threadNum,cyclicBarrier).testTime();          new LockTest("Lock",round,threadNum,cyclicBarrier).testTime();          new AtomicTest("Atom",round,threadNum,cyclicBarrier).testTime();      }        public static void main(String args[]){                    for(int i=0;i<5;i++){              int round=100000*(i+1);              int threadNum=5*(i+1);              CyclicBarrier cb=new CyclicBarrier(threadNum*2+1);              out.PRintln("==========================");              out.println("round:"+round+" thread:"+threadNum);              test(round,threadNum,cb);                        }      }  }    class SyncTest extends TestTemplate{      public SyncTest(String _id,int _round,int _threadNum,CyclicBarrier _cb){          super( _id, _round, _threadNum, _cb);      }      @Override      /**      * synchronized关键字不在方法签名里面,所以不涉及重载问题      */      synchronized long  getValue() {          return super.countValue;      }      @Override      synchronized void  sumValue() {          super.countValue+=preInit[index++%round];      }  }      class LockTest extends TestTemplate{      ReentrantLock lock=new ReentrantLock();      public LockTest(String _id,int _round,int _threadNum,CyclicBarrier _cb){          super( _id, _round, _threadNum, _cb);      }      /**      * synchronized关键字不在方法签名里面,所以不涉及重载问题      */      @Override      long getValue() {          try{              lock.lock();              return super.countValue;          }finally{              lock.unlock();          }      }      @Override      void sumValue() {          try{              lock.lock();              super.countValue+=preInit[index++%round];          }finally{              lock.unlock();          }      }  }      class AtomicTest extends TestTemplate{      public AtomicTest(String _id,int _round,int _threadNum,CyclicBarrier _cb){          super( _id, _round, _threadNum, _cb);      }      @Override      /**      * synchronized关键字不在方法签名里面,所以不涉及重载问题      */      long  getValue() {          return super.countValueAtmoic.get();      }      @Override      void  sumValue() {          super.countValueAtmoic.addAndGet(super.preInit[indexAtomic.get()%round]);      }  }  abstract class TestTemplate{      private String id;      protected int round;      private int threadNum;      protected long countValue;      protected AtomicLong countValueAtmoic=new AtomicLong(0);      protected int[] preInit;      protected int index;      protected AtomicInteger indexAtomic=new AtomicInteger(0);      Random r=new Random(47);      //任务栅栏,同批任务,先到达wait的任务挂起,一直等到全部任务到达制定的wait地点后,才能全部唤醒,继续执行      private CyclicBarrier cb;      public TestTemplate(String _id,int _round,int _threadNum,CyclicBarrier _cb){          this.id=_id;          this.round=_round;          this.threadNum=_threadNum;          cb=_cb;          preInit=new int[round];          for(int i=0;i<preInit.length;i++){              preInit[i]=r.nextInt(100);          }      }            abstract void sumValue();      /*      * 对long的操作是非原子的,原子操作只针对32位      * long是64位,底层操作的时候分2个32位读写,因此不是线程安全      */      abstract long getValue();        public void testTime(){          ExecutorService se=Executors.newCachedThreadPool();          long start=System.nanoTime();          //同时开启2*ThreadNum个数的读写线程          for(int i=0;i<threadNum;i++){              se.execute(new Runnable(){                  public void run() {                      for(int i=0;i<round;i++){                          sumValue();                      }                        //每个线程执行完同步方法后就等待                      try {                          cb.await();                      } catch (InterruptedException e) {                          // TODO Auto-generated catch block                          e.printStackTrace();                      } catch (BrokenBarrierException e) {                          // TODO Auto-generated catch block                          e.printStackTrace();                      }                      }              });              se.execute(new Runnable(){                  public void run() {                        getValue();                      try {                          //每个线程执行完同步方法后就等待                          cb.await();                      } catch (InterruptedException e) {                          // TODO Auto-generated catch block                          e.printStackTrace();                      } catch (BrokenBarrierException e) {                          // TODO Auto-generated catch block                          e.printStackTrace();                      }                    }              });          }                    try {              //当前统计线程也wait,所以CyclicBarrier的初始值是threadNum*2+1              cb.await();          } catch (InterruptedException e) {              // TODO Auto-generated catch block              e.printStackTrace();          } catch (BrokenBarrierException e) {              // TODO Auto-generated catch block              e.printStackTrace();          }          //所有线程执行完成之后,才会跑到这一步          long duration=System.nanoTime()-start;          out.println(id+" = "+duration);                }    }  
上一篇:2.4

下一篇:文件夹和文件压缩

发表评论 共有条评论
用户名: 密码:
验证码: 匿名发表