Java8 更快的原子类:LongAdder(笔记)

• 更快的原子类:LongAdder

     大家对AtomicInteger的基本实现机制应该比较了解,它们是在一个死循环内,不断尝试修改目标值,知道修改成功,如果竞争不激烈,那么修改成功的概率就很高,否则,修改失败的概率就很高,在大量修改失败时,这些原子操作就会进行多次循环尝试,因此性能就会受到影响

     那么竞争激烈的时候,我们应该如何进一步提高系统性能呢?一种基本方案就是可以使用热点分离,将竞争的数据进行分解.基于这个思路,打击应该可以想到一种对传统AtomicInteger等原子类的改进方法,虽然在CAS操作中没有锁,但是像减少锁粒度这种分离热点的思路依然可以使用,一种可行的方案就是仿造ConcurrengHashMap,将热点数据分离,比如,可以将AtomicInteger的内部核心数据value分离成一个数组,每个线程访问时,通过哈希等算法映射到其中一个数字进行计数,而最终的计数结果,则为这个数组的求和累加,其中,热点数据value被分离成多个单元cell,每个cell独自维护内部的值,当前对象的实际值由所有的cell累计合成,这样,热点就进行了有效的分离,提高了并行度,LongAdder正是使用了这种思想.

• 测试LongAdder,原子类以及同步锁性能测试

• public class LongAdderDemo {
      private static final int MAX_THREADS = 3;
      private static final int TASK_COUNT = 3;
      private static final int TARGET_COUNT = 10000000;
  
      private AtomicLong acount = new AtomicLong(0L);
      private LongAdder lacount = new LongAdder();
      private long count = 0;
  
  
      private static CountDownLatch cdlsync = new CountDownLatch(TASK_COUNT);
      private static CountDownLatch cdlatomic = new CountDownLatch(TASK_COUNT);
      private static CountDownLatch cdladdr = new CountDownLatch(TASK_COUNT);
  
  
      protected synchronized long inc() {
          return ++count;
      }
  
      protected synchronized long getCount() {
          return count;
      }
  
      public class SyncThread implements Runnable {
          protected String name;
          protected long starttime;
          LongAdderDemo out;
  
          public SyncThread(long starttime, LongAdderDemo out) {
              this.starttime = starttime;
              this.out = out;
          }
  
          @Override
          public void run() {
              long v = out.getCount();
              while (v < TARGET_COUNT) {
                  v = out.inc();
              }
              long endtime = System.currentTimeMillis();
              System.out.println("SyncThread spend:" + (endtime - starttime) + "ms" + " v" + v);
              cdlsync.countDown();
          }
      }
  
      public void testSync() throws InterruptedException {
          ExecutorService exe = Executors.newFixedThreadPool(MAX_THREADS);
          long starttime = System.currentTimeMillis();
          SyncThread sync = new SyncThread(starttime, this);
          for (int i = 0; i < TASK_COUNT; i++) {
              exe.submit(sync);
          }
          cdlsync.await();
          exe.shutdown();
      }
  
      public class AtomicThread implements Runnable {
          protected String name;
          protected long starttime;
  
          public AtomicThread(long starttime) {
              this.starttime = starttime;
          }
  
          @Override
          public void run() {
              long v = acount.get();
              while (v < TARGET_COUNT) {
                  v = acount.incrementAndGet();
              }
              long endtime = System.currentTimeMillis();
              System.out.println("AtomicThread spend:" + (endtime - starttime) + "ms" + " v" + v);
              cdlatomic.countDown();
          }
      }
  
      public void testAtomic() throws InterruptedException {
          ExecutorService exe = Executors.newFixedThreadPool(MAX_THREADS);
          long starttime = System.currentTimeMillis();
          AtomicThread atomic = new AtomicThread(starttime);
          for (int i = 0; i < TASK_COUNT; i++) {
              exe.submit(atomic);
          }
          cdlatomic.await();
          exe.shutdown();
      }
  
      public class LongAdderThread implements Runnable {
          protected String name;
          protected long starttime;
  
          public LongAdderThread(long starttime) {
              this.starttime = starttime;
          }
  
          @Override
          public void run() {
              long v = lacount.sum();
              while (v < TARGET_COUNT) {
                  lacount.increment();
                  v = lacount.sum();
              }
              long endtime = System.currentTimeMillis();
              System.out.println("LongAdderThread spend:" + (endtime - starttime) + "ms" + " v" + v);
              cdladdr.countDown();
          }
  
      }
  
      public void testLongAdder() throws InterruptedException {
          ExecutorService exe = Executors.newFixedThreadPool(MAX_THREADS);
          long starttime = System.currentTimeMillis();
          LongAdderThread atomic = new LongAdderThread(starttime);
          for (int i = 0; i < TASK_COUNT; i++) {
              exe.submit(atomic);
          }
          cdladdr.await();
          exe.shutdown();
      }
  
      public static void main(String[] args) throws InterruptedException {
          LongAdderDemo demo = new LongAdderDemo();
          demo.testSync();
          demo.testAtomic();
          demo.testLongAdder();
      }
  }

  结果：

• SyncThread spend:317ms v10000000
  SyncThread spend:317ms v10000001
  SyncThread spend:317ms v10000002
  AtomicThread spend:157ms v10000000
  AtomicThread spend:160ms v10000001
  AtomicThread spend:179ms v10000002
  LongAdderThread spend:202ms v10000001
  LongAdderThread spend:202ms v10000001
  LongAdderThread spend:203ms v10000001

  SyncThread spend:535ms v10000000
  SyncThread spend:536ms v10000001
  SyncThread spend:542ms v10000002
  AtomicThread spend:184ms v10000000
  AtomicThread spend:185ms v10000001
  AtomicThread spend:214ms v10000002
  LongAdderThread spend:209ms v10000000
  LongAdderThread spend:209ms v10000000
  LongAdderThread spend:212ms v10000001
  

  SyncThread spend:280ms v10000000
  SyncThread spend:280ms v10000001
  SyncThread spend:281ms v10000002
  AtomicThread spend:102ms v10000001
  AtomicThread spend:102ms v10000000
  AtomicThread spend:106ms v10000001
  LongAdderThread spend:209ms v10000000
  LongAdderThread spend:209ms v10000000
  LongAdderThread spend:210ms v10000000
  

  SyncThread spend:391ms v10000002
  SyncThread spend:391ms v10000001
  SyncThread spend:391ms v10000000
  AtomicThread spend:185ms v10000000
  AtomicThread spend:185ms v10000001
  AtomicThread spend:188ms v10000002
  LongAdderThread spend:210ms v10000000
  LongAdderThread spend:212ms v10000001
  LongAdderThread spend:215ms v10000002