关于synchronized与lock的性能比较

记得当初看教程的时候大家都说lock性能比好不少，最近需要自己设计一个缓存终于要自己尝试一番了。

1.关于两者的实现的比较

A).一般认为synchronized关键字的实现是源自于像信号量之类的线程同步机制，涉及到线程运行状态的切换，在高并发状态下，CPU消耗过多的时间在线程的调度上，从而造成了性能的极大浪费。然而真的如此么？
B).lock实现原理则是依赖于硬件，现代处理器都支持CAS指令，所谓CAS指令简单的来说Compare And Set，CPU循环执行指令直到得到所期望的结果，换句话来说就是当变量真实值不等于当前线程调用时的值的时候(说明其他线程已经将这个值改变)，就不会赋予变量新的值。这样就保证了变量在多线程环境下的安全性。

然而，现实情况是当JDK版本高于1.6的时候，synchronized已经被做了CAS的优化：具体是这样的，当执行到synchronized代码块时，先对对象头的锁标志位用lock cmpxchg的方式设置成“锁住“状态，释放锁时，在用lock cmpxchg的方式修改对象头的锁标志位为”释放“状态，写操作都立刻写回主内存。JVM会进一步对synchronized时CAS失败的那些线程进行阻塞操作(调用操作系统的信号量)(此段来摘自别处)。也就是先CAS操作，不行的话继而阻塞线程。

除此之外，系统环境，CPU架构，虚拟机环境都会影响两者的性能关系。

2.用数据说话

1).X86_64 cpu i7 4910mq @4.0ghz ,Windows10 64bit,JDK1.8 hotspot 64bit虚拟机环境

测试代码

测试对某Map对象高并发下的读写线程安全测试
测试对比有synchronized，ReadWriteLock，ConcurrentHashMap，

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49

public class MapTest {        private Map<integer,string> map = new ConcurrentHashMap<>();        private long starttime;        private AtomicInteger count = new AtomicInteger(t_count);        private final static int t_count = 5000;        private final static int rw_count = 10000;           Runnable readrun = new Runnable() {             @Override             public void run() {                 int i = rw_count;                 while (i > 0){                     map.get(i);                     i--;                 }                 System.out.println("read-mapsize="+map.size());                 if(count.decrementAndGet() == 0)                     System.out.println("time="+ (System.currentTimeMillis() - starttime +"ms"));             }         };           Runnable writerun = new Runnable() {             @Override             public void run() {                 int i = rw_count;                 while (i > 0){                     map.put(i,i+"");                     i--;                 }                 System.out.println("write-mapsize="+map.size());                 if(count.decrementAndGet() == 0)                     System.out.println("time="+ (System.currentTimeMillis() - starttime + "ms"));             }         };           public void run(){             starttime = System.currentTimeMillis();             for(int i = 0;i < t_count/2;i ++){                 new Thread(writerun).start();                 new Thread(readrun).start();             }         } } </integer,string>

HashMap 用synchronized重写

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

public class SyncHashMap extends HashMap{       @Override     public Object get(Object key) {         // TODO Auto-generated method stub         synchronized (this) {             return super.get(key);         }     }       @Override     public synchronized Object put(Object key, Object value) {         // TODO Auto-generated method stub         synchronized (this) {             return super.put(key, value);         }       }   }

用读写锁实现的Map代理类，有些粗糙，没加try finally

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103

public class SyncMapProxy<k,v> implements Map<k,v>{       private Map<k,v> origin;     private ReadWriteLock lock;       public SyncMapProxy(Map<k, v=""> origin) {         this.origin = origin;         lock = new ReentrantReadWriteLock();     }       public static  <k,v> SyncMapProxy<k,v> SyncMap(Map<k,v> map){         return new SyncMapProxy<k,v>(map);     }       @Override     public void clear() {         lock.writeLock().lock();         origin.clear();         lock.writeLock().unlock();     }       @Override     public boolean containsKey(Object key) {         lock.readLock().lock();         boolean res = origin.containsKey(key);         lock.readLock().unlock();         return res;     }       @Override     public boolean containsValue(Object value) {         lock.readLock().lock();         boolean res = origin.containsKey(value);         lock.readLock().unlock();         return res;     }       @Override     public Set<entry<k, v="">> entrySet() {         lock.readLock().lock();         Set<entry<k, v="">> res = origin.entrySet();         lock.readLock().unlock();         return res;     }       @Override     public V get(Object key) {         lock.readLock().lock();         V res = origin.get(key);         lock.readLock().unlock();         return res;     }       @Override     public boolean isEmpty() {         return origin.isEmpty();     }       @Override     public Set<k> keySet() {         lock.readLock().lock();         Set<k> res = origin.keySet();         lock.readLock().unlock();         return res;     }       @Override     public V put(K key, V value) {         lock.writeLock().lock();         V v = origin.put(key, value);         lock.writeLock().unlock();         return v;     }       @Override     public void putAll(Map<!--? extends K, ? extends V--> map) {         lock.writeLock().lock();         origin.putAll(map);         lock.writeLock().unlock();     }       @Override     public V remove(Object key) {         lock.writeLock().lock();         V v = origin.remove(key);         lock.writeLock().unlock();         return v;     }       @Override     public int size() {         return origin.size();     }       @Override     public Collection<v> values() {         lock.readLock().lock();         Collection<v> res = origin.values();         lock.readLock().unlock();         return res;     } } </v></v></k></k></entry<k,></entry<k,></k,v></k,v></k,v></k,v></k,></k,v></k,v></k,v>

并发量100000,每个线程对Map执行读写100次,总耗时
ConcurrentHashMap:6112ms
synchronized:6121ms
ReadWriteLock:6182ms
Collections.synchronizedMap:6175ms

并发量10000,每个线程对Map执行读写1000次,总耗时
ConcurrentHashMap:1126ms
synchronized:1145ms
ReadWriteLock:2086ms
Collections.synchronizedMap:1170ms

并发量5000,每个线程对Map执行读写10000次,总耗时
ConcurrentHashMap:1206ms
synchronized:4896ms
ReadWriteLock:8505ms
Collections.synchronizedMap:4883ms

并发量1000,每个线程对Map执行读写100000次,总耗时
ConcurrentHashMap:1748ms
synchronized:9341ms
ReadWriteLock:18720ms
Collections.synchronizedMap:8945ms

并发量100,每个线程对Map执行读写1000000次,总耗时
ConcurrentHashMap:1922ms
synchronized:8417ms
ReadWriteLock:16110ms
Collections.synchronizedMap:9604ms

事实证明在以上的配置环境JDK1.8 X86 Windows10下，高并发下这几种方式性能都相差无几，较高和较低并发下，synchronized都比ReadWriteLock来的快，基本是两倍的关系。ConcurrentHashMap作为同步的Map还是时间性能还是最高的。总之在hotspot下都是一个数量级的。

---

2).下面看另外一种环境
Android6.0 X86_64模拟器镜像，ART Runtime

并发量20,每个线程对Map执行读写1000000次,总耗时
ConcurrentHashMap:10841ms
synchronized:239452ms
ReadWriteLock:16450ms
Collections.synchronized:213429ms

并发量200,每个线程对Map执行读写10000次,总耗时
ConcurrentHashMap:973ms
synchronized:57047ms
ReadWriteLock:1274ms
Collections.synchronized:52746ms

**难以置信的性能差距，synchronized和Lock在Android的Art环境下确实有着一个数量级的差距，可达数十倍之多，但是在Hotspot环境下却恰恰相反，lock在多数情况下反而不如synchronized。
这里估计是Android Art虚拟机尚未对synchronized进行CAS优化，主要还是因为Android现在作为客户端操作系统，对高并发的资源竞争并无必要做优化，以上结果尚不能下定论，看来要去扒一扒Art的源码才能知道具体的原因了。**

总结

如果开发的是服务器程序，并且使用的是最新的hotspot虚拟机，synchronized和lock其实已经相差无几，其底层实现已经差不多了。但是如果你是Android开发者，使用synchronized还是需要考虑其性能差距的。