ReentrantLock 在多線程情況下要遠勝synchronize,這點沒有疑問。
最近要寫個程式,有個變數是有多數情況下是一個線程讀寫,有少數情況下是多個線程並發讀寫。所以要測試下ReentrantLock 在單線程下和synchronize的效率對比。
在測試的過程中發現一個有意思的現象。
測試代碼見後面。
測試代碼1結果:
noLockTime: 0:00:00.004
noLockTime: 0:00:00.006
noLockTime: 0:00:00.000
syncTime: 0:00:02.902
syncTime: 0:00:02.902
syncTime: 0:00:02.886
ReentrantTime: 0:00:05.012
ReentrantTime: 0:00:05.007
ReentrantTime: 0:00:04.270
則開始看到這個測試結果,覺得相當奇怪,ReentrantLock居然比synchronize要慢這麼多。
但是後來再仔細研究測試結果,發現原來是jvm把測試代碼最佳化掉了!觀察noLockTime的時間,差不多是0!
於是加上total的統計代碼,得到以下的測試結果:
加上total統計的測試代碼結果:
noLockTime: 0:00:03.786
total:1078760960
noLockTime: 0:00:03.774
total:1078760960
noLockTime: 0:00:03.428
total:1078760960
syncTime: 0:00:05.553
total:1078760960
syncTime: 0:00:05.555
total:1078760960
syncTime: 0:00:04.826
total:1078760960
ReentrantTime: 0:00:05.451
total:1078760960
ReentrantTime: 0:00:05.424
total:1078760960
ReentrantTime: 0:00:04.493
total:1078760960
這個結果算是正常的結果了。
總結:
- 以前一直聽說JVM可以在運行時最佳化代碼,果然是相當的牛B的功能,直接在運行時把無用的代碼全最佳化掉了。
- 測試程式要注意防止jvm等的最佳化
- ReentrantLock和synchronize在單線程下效率基本一樣(從實現原理來看,在單線程情況下,都是一個比較指令即可)
- 在單線程下,ReentrantLock和synchronize的效率是非常高的,一次調用約10e-11秒。
測試代碼1:
import java.util.concurrent.locks.ReentrantLock;import org.apache.commons.lang3.time.StopWatch;public class XXX {static ReentrantLock lock = new ReentrantLock(false);static int size = 100;static int count = 100000000;static StopWatch watch = new StopWatch();public static void main(String[] args) {noLockTime();noLockTime();noLockTime();syncTime();syncTime();syncTime();ReentrantTime();ReentrantTime();ReentrantTime();}static void noLockTime(){//int total = 0;watch.reset();watch.start();for(int i = 0; i < count; ++i){testNotLock();//total += testNotLock();}watch.stop();System.out.println("noLockTime: " + watch);//System.out.println("total:" + total);}static void syncTime(){//int total = 0;watch.reset();watch.start();for(int i = 0; i < count; ++i){testSync();//total += testSync();}watch.stop();System.out.println("syncTime: " + watch);//System.out.println("total:" + total);}static void ReentrantTime(){//int total = 0;watch.reset();watch.start();for(int i = 0; i < count; ++i){testReentrantLock();//total += testReentrantLock();}watch.stop();System.out.println("ReentrantTime: " + watch);//System.out.println("total:" + total);}static int testNotLock(){int sum = 0;for(int i = 0; i < size; ++i){sum += i;}return sum;}static synchronized int testSync(){int sum = 0;for(int i = 0; i < size; ++i){sum += i;}return sum;}static int testReentrantLock(){try{lock.lock();int sum = 0;for(int i = 0; i < size; ++i){sum += i;}return sum;}finally{lock.unlock();}}}
加上total統計的測試代碼:
import java.util.concurrent.locks.ReentrantLock;import org.apache.commons.lang3.time.StopWatch;public class XXX {static ReentrantLock lock = new ReentrantLock(false);static int size = 100;static int count = 100000000;static StopWatch watch = new StopWatch();public static void main(String[] args) {noLockTime();noLockTime();noLockTime();syncTime();syncTime();syncTime();ReentrantTime();ReentrantTime();ReentrantTime();}static void noLockTime(){int total = 0;watch.reset();watch.start();for(int i = 0; i < count; ++i){//testNotLock();total += testNotLock();}watch.stop();System.out.println("noLockTime: " + watch);System.out.println("total:" + total);}static void syncTime(){int total = 0;watch.reset();watch.start();for(int i = 0; i < count; ++i){//testSync();total += testSync();}watch.stop();System.out.println("syncTime: " + watch);System.out.println("total:" + total);}static void ReentrantTime(){int total = 0;watch.reset();watch.start();for(int i = 0; i < count; ++i){//testReentrantLock();total += testReentrantLock();}watch.stop();System.out.println("ReentrantTime: " + watch);System.out.println("total:" + total);}static int testNotLock(){int sum = 0;for(int i = 0; i < size; ++i){sum += i;}return sum;}static synchronized int testSync(){int sum = 0;for(int i = 0; i < size; ++i){sum += i;}return sum;}static int testReentrantLock(){try{lock.lock();int sum = 0;for(int i = 0; i < size; ++i){sum += i;}return sum;}finally{lock.unlock();}}}