ForkJoin框架与读写锁
Fork/Join框架就是在必要的情况下,将一个大任务,进行拆分(fork)成若千个小任务(拆到不可再拆时),再将一个个的小任务运算的结果进行join汇总。
ForkJoin框架采用工作窃取
模式(work-stealing) :当执行新的任务时它可以将其拆分分成更小的任务执行,并将小任务加到线程队列中,然后再从一个随机线程的队列中偷一个并把它放在自己的队列中。
相对于一般的线程池实现,fork/join框架的优势体现在对其中包含的任务的处理方式上,在一般的线程池中,如果一个线程正在执行的任务由于某些原因无法继续运行,那么该线程会处于等待状态。而在fork/join框架实现中,如果某个子问题由于等待另外一个子问题的完成而无法继续运行。那么处理该子问题的线程会主动寻找其他尚未运行的子问题来执行。这种方式减少了线程的等待时间,提高了性能。
ForkJoin框架
下面是一个很简单的示例,即用Fork/Join框架来计算0-500亿的和,普通For用时13745毫秒,Fork/Join框架用时8846毫秒,而且还有拆装箱的时间,足以看出Fork/Join框架的优势。
1import java.time.Duration;
2import java.time.Instant;
3import java.util.concurrent.ForkJoinPool;
4import java.util.concurrent.ForkJoinTask;
5import java.util.concurrent.RecursiveTask;
6
7public class TestForkJoinPool {
8 public static void main(String[] args) {
9 Instant start = Instant.now();
10 ForkJoinPool forkJoinPool = new ForkJoinPool();
11 ForkJoinTask<Long> task = new ForkJoinSunCalculate(0L, 50000000000L);
12 Long invoke = forkJoinPool.invoke(task);
13 System.out.println(invoke);
14 Instant end = Instant.now();
15 System.out.println(Duration.between(start, end).toMillis()); //8846
16 }
17
18 public static void main(String[] args) {
19 Instant start = Instant.now();
20 long sum = 0L;
21 for (long i = 0; i < 50000000000L; i++) {
22 sum += i;
23 }
24 System.out.println(sum);
25 Instant end = Instant.now();
26 System.out.println(Duration.between(start, end).toMillis()); //13745
27 }
28}
29
30class ForkJoinSunCalculate extends RecursiveTask<Long> {
31 private long start;
32 private long end;
33
34 //临界值
35 private static final long VALUE = 10000L;
36
37 public ForkJoinSunCalculate(long start, long end){
38 this.start = start;
39 this.end = end;
40 }
41
42 @Override
43 protected Long compute() {
44 long length = end - start;
45 if(length <= VALUE){
46 long sum = 0L;
47 for (long i = start; i <= end; i++) {
48 sum += i;
49 }
50 return sum;
51 }else{
52 long middle = (end - start) / 2 + start;
53 ForkJoinSunCalculate leftCalculate = new ForkJoinSunCalculate(start, middle);
54 leftCalculate.fork(); //进行拆分,同时压入线程队列
55
56 ForkJoinSunCalculate rightCalculate = new ForkJoinSunCalculate(middle + 1, end);
57 rightCalculate.fork();//进行拆分,同时压入线程队列
58
59 return leftCalculate.join() + rightCalculate.join();
60 }
61 }
62}
用JDK8的特性stream谁快呢?
1import java.time.Duration;
2import java.time.Instant;
3import java.util.concurrent.ForkJoinPool;
4import java.util.concurrent.ForkJoinTask;
5import java.util.concurrent.RecursiveTask;
6import java.util.stream.LongStream;
7
8public class TestForkJoinPool {
9 public static void main(String[] args) {
10 Instant start = Instant.now();
11 long sum = LongStream.rangeClosed(0L, 50000000000L)
12 .parallel()
13 .reduce(0L, Long::sum);
14 System.out.println(sum);
15 Instant end = Instant.now();
16 System.out.println(Duration.between(start, end).toMillis()); //5156
17 }
18}
其实这简单的计算还是stream更快,底层优化太多了。
ReadWriteLock
ReadWriteLock维护了一对相关的锁,一个用于只读操作,另一个用于写入操作。只要没有writer,读取锁可以由多个reader线程同时保持,写入锁是独占的。
ReadWriteLock读取操作通常不会改变共享资源,但执行写入操作时,必须独占方式来获取锁。对于读取操作占多数的数据结构。ReadWriteLock 能提供比独占锁更高的并发性。而对于只读的数据结构,其中包含的不变性可以完全不需要考虑加锁操作。
1package thread_study;
2
3import java.util.concurrent.locks.ReadWriteLock;
4import java.util.concurrent.locks.ReentrantReadWriteLock;
5
6//读写锁
7public class TestReadWriteLock {
8 public static void main(String[] args) {
9 ReadWriteLockDemo writeLockDemo = new ReadWriteLockDemo();
10 new Thread(()->{
11 for (int i = 0; i < 30; i++) {
12 writeLockDemo.set(i);
13 }
14 }, "Write Thread").start();
15 for (int i = 0; i < 20; i++) {
16 new Thread(writeLockDemo::get, "Read Thread").start();
17 }
18 }
19}
20
21class ReadWriteLockDemo{
22 private int number = 0;
23 private ReadWriteLock lock = new ReentrantReadWriteLock();
24
25 public void get(){
26 lock.readLock().lock();
27 try{
28 System.out.println(Thread.currentThread().getName() + " : " + number);
29 }finally {
30 lock.readLock().unlock();
31 }
32 }
33
34 public void set(int number){
35 lock.writeLock().lock();
36 try{
37 System.out.println(Thread.currentThread().getName());
38 this.number = number;
39 }finally {
40 lock.writeLock().unlock();
41 }
42 }
43}