转自http://www.iteye.com/topic/333669
JDK1.5以后加入了concurrent包,主要是为了提高多线程的开发效率,其中提供了很多支持并发的集合类,其中包括:ConcurrentHashMap。大家知道HashTable也是支持并发环境的,也就是说多线程安全的,那两者有什么区别呢?
分析
其实简单的说是同步机制有区别,具体区别又在那里呢?
请看HashTable的put方法:
- /**
- * Maps the specified <code>key</code> to the specified
- * <code>value</code> in this hashtable. Neither the key nor the
- * value can be <code>null</code>. <p>
- *
- * The value can be retrieved by calling the <code>get</code> method
- * with a key that is equal to the original key.
- *
- * @param key the hashtable key
- * @param value the value
- * @return the previous value of the specified key in this hashtable,
- * or <code>null</code> if it did not have one
- * @exception NullPointerException if the key or value is
- * <code>null</code>
- * @see Object#equals(Object)
- * @see #get(Object)
- */
- public synchronized V put(K key, V value) {
- // Make sure the value is not null
- if (value == null) {
- throw new NullPointerException();
- }
- // Makes sure the key is not already in the hashtable.
- Entry tab[] = table;
- int hash = key.hashCode();
- int index = (hash & 0x7FFFFFFF) % tab.length;
- for (Entry<K,V> e = tab[index] ; e != null ; e = e.next) {
- if ((e.hash == hash) && e.key.equals(key)) {
- V old = e.value;
- e.value = value;
- return old;
- }
- }
- modCount++;
- if (count >= threshold) {
- // Rehash the table if the threshold is exceeded
- rehash();
- tab = table;
- index = (hash & 0x7FFFFFFF) % tab.length;
- }
- // Creates the new entry.
- Entry<K,V> e = tab[index];
- tab[index] = new Entry<K,V>(hash, key, value, e);
- count++;
- return null;
- }
代码中使用synchronized函数的方式进行同步,这个类的其他方法也是使用这个机制进行同步的。我们先来大致了解一下synchronized的机制:
在多线程环境中,Java的对象都会隐式的包含有一个同步队列,其中类会有一个,然后每个类实例也会包含一个。如图:
- class Foo {
- synchronized void doSomething(); // 把同步的线程放入类实例的同步队列
- synchronized static void doSomething(); //把同步的线程放入类的同步队列
- }
然后我们再来看看ConcurrentHashMap 的put方法:
- //ConcurrentHashMap
- public V put(K key, V value) {
- if (value == null)
- throw new NullPointerException();
- int hash = hash(key.hashCode());
- return segmentFor(hash).put(key, hash, value, false);
- }
- //Segment内部类,继承至ReentrantLock
- V put(K key, int hash, V value, boolean onlyIfAbsent) {
- lock();
- try {
- int c = count;
- if (c++ > threshold) // ensure capacity
- rehash();
- HashEntry<K,V>[] tab = table;
- int index = hash & (tab.length - 1);
- HashEntry<K,V> first = tab[index];
- HashEntry<K,V> e = first;
- while (e != null && (e.hash != hash || !key.equals(e.key)))
- e = e.next;
- V oldValue;
- if (e != null) {
- oldValue = e.value;
- if (!onlyIfAbsent)
- e.value = value;
- }
- else {
- oldValue = null;
- ++modCount;
- tab[index] = new HashEntry<K,V>(key, hash, first, value);
- count = c; // write-volatile
- }
- return oldValue;
- } finally {
- unlock();
- }
- }
ReentrantLock就是Java Concurrent包提供的锁对象,Lock的使用方法大致如下:
- Lock l = ...;
- l.lock();
- try {
- // access the resource protected by this lock
- } finally {
- l.unlock();
- }
为什么使用Lock对象会比使用synchronized有更好的性能呢?我们再来看看ReentrantLock的实现:
- public class ReentrantLock implements Lock, java.io.Serializable {
- private static final long serialVersionUID = 7373984872572414699L;
- /** Synchronizer providing all implementation mechanics */
- private final Sync sync;
- /**
- * Base of synchronization control for this lock. Subclassed
- * into fair and nonfair versions below. Uses AQS state to
- * represent the number of holds on the lock.
- */
- static abstract class Sync extends AbstractQueuedSynchronizer {
- private static final long serialVersionUID = -5179523762034025860L;
- ..........
- ...........
我们从中看到ReentrantLock对象也维护着一个同步队列,性能差别就在于这个队列的实现上,我们再来看AbstractQueuedSynchronizer的代码:
- /**
- * Inserts node into queue, initializing if necessary. See picture above.
- * @param node the node to insert
- * @return node's predecessor
- */
- private Node enq(final Node node) {
- for (;;) {
- Node t = tail;
- if (t == null) { // Must initialize
- Node h = new Node(); // Dummy header
- h.next = node;
- node.prev = h;
- if (compareAndSetHead(h)) {
- tail = node;
- return h;
- }
- }
- else {
- node.prev = t;
- if (compareAndSetTail(t, node)) {
- t.next = node;
- return t;
- }
- }
- }
- }
- /**
- * CAS head field. Used only by enq
- */
- private final boolean compareAndSetHead(Node update) {
- return unsafe.compareAndSwapObject(this, headOffset, null, update);//使用compare and swap方式
- }
- /**
- * CAS tail field. Used only by enq
- */
- private final boolean compareAndSetTail(Node expect, Node update) {
- return unsafe.compareAndSwapObject(this, tailOffset, expect, update);//使用compare and swap方式
- }
到了Unsafe.java这里就要通过本地代码实现了,下面是kaffe里面的本地代码实现:
- /**
- * Helper macro, defining a sun.misc.Unsafe compare and swap function
- * with a given NAME tail and TYPE of arguments.
- */
- #define KAFFE_UNSAFE_COMPARE_AND_SWAP(NAME, TYPE)
- JNIEXPORT jboolean JNICALL Java_sun_misc_Unsafe_compareAndSwap ## NAME(JNIEnv* env, jobject unsafe UNUSED, jobject obj, jlong offset, TYPE expect, TYPE update)
- {
- volatile TYPE * address = getFieldAddress(env, obj, offset);
- if (sizeof(TYPE) == sizeof(gint))
- return g_atomic_int_compare_and_exchange((volatile gint *) address, (gint) expect, (gint) update);
- else if (sizeof(TYPE) == sizeof(gpointer))
- return g_atomic_pointer_compare_and_exchange((volatile gpointer *) address, (gpointer) expect, (gpointer) update);
- else
- if (*address == expect) {
- *address = update;
- return JNI_TRUE;
- }
- else
- return JNI_FALSE;
- }
再看glib的代码:
- gboolean g_atomic_int_compare_and_exchange (volatile gint *atomic,
- gint oldval,
- gint newval)
- {
- gint result;
- __asm__ __volatile__ ("lock; cmpxchgl %2, %1"
- : "=a" (result), "=m" (*atomic)
- : "r" (newval), "m" (*atomic), "0" (oldval));
- return result == oldval;
- }
AT&T汇编,
一条指令总会是原子性的了吧
总结
concurrent包中大量使用了新的锁机制,新的Lock机制最终归结到一个原子性操作上,所以会比使用synchronized关键字有更高的性能。