5.0的多線程任務包對於同步的性能方面有了很大的改進,在原有synchronized關鍵字的基礎上,又增加了ReentrantLock,以及各種Atomic類。了解其性能的優劣程度,有助與我們在特定的情形下做出正確的選擇。
總體的結論先擺出來:
synchronized:
在資源競爭不是很激烈的情況下,偶爾會有同步的情形下,synchronized是很合適的。原因在於,編譯程序通常會盡可能的進行優化synchronize,另外可讀性非常好,不管用沒用過5.0多線程包的程序員都能理解。
ReentrantLock:
ReentrantLock提供了多樣化的同步,比如有時間限制的同步,可以被Interrupt的同步(synchronized的同步是不能Interrupt的)等。在資源競爭不激烈的情形下,性能稍微比synchronized差點點。但是當同步非常激烈的時候,synchronized的性能一下子能下降好幾十倍。而ReentrantLock確還能維持常態。
Atomic:
和上面的類似,不激烈情況下,性能比synchronized略遜,而激烈的時候,也能維持常態。激烈的時候,Atomic的性能會優於ReentrantLock一倍左右。但是其有一個缺點,就是只能同步一個值,一段代碼中只能出現一個Atomic的變量,多於一個同步無效。因為他不能在多個Atomic之間同步。
所以,我們寫同步的時候,優先考慮synchronized,如果有特殊需要,再進一步優化。ReentrantLock和Atomic如果用的不好,不僅不能提高性能,還可能帶來災難。
先貼測試結果:再貼代碼(Atomic測試代碼不准確,一個同步中只能有1個Actomic,這里用了2個,但是這里的測試只看速度)
==========================
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代碼
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關鍵字不在方法簽名里面,所以不涉及重載問題 */ 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關鍵字不在方法簽名里面,所以不涉及重載問題 */ @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關鍵字不在方法簽名里面,所以不涉及重載問題 */ 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); //任務柵欄,同批任務,先到達wait的任務掛起,一直等到全部任務到達制定的wait地點后,才能全部喚醒,繼續執行 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(); /* * 對long的操作是非原子的,原子操作只針對32位 * long是64位,底層操作的時候分2個32位讀寫,因此不是線程安全 */ abstract long getValue(); public void testTime(){ ExecutorService se=Executors.newCachedThreadPool(); long start=System.nanoTime(); //同時開啟2*ThreadNum個數的讀寫線程 for(int i=0;i<threadNum;i++){ se.execute(new Runnable(){ public void run() { for(int i=0;i<round;i++){ sumValue(); } //每個線程執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 { //每個線程執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 { //當前統計線程也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(); } //所有線程執行完成之后,才會跑到這一步 long duration=System.nanoTime()-start; out.println(id+" = "+duration); } }