我正在尝试各种使大型计算并行运行的方法。除明确地创建和启动Thread
实例(这要慢得多)之外,它的工作符合预期。我添加了一些println输出,以检查线程是否真正并行运行,但所做的更改使它快得多。速度从“超过一秒”变为“数十毫秒”,结果是可重复的。
非常古怪:ArrayList
没有慢度问题,并且println
没有明显的区别,但是LinkedList
直到我添加一些println
为止都很慢。
我在具有Java 8的Windows 10上运行此程序。我还尝试了Java 11和12,并且没有发生速度慢(因此无意义的输出无法使其更快)的情况。
我的问题不是如何使其更快-我实际上不需要运行此代码,而我可以用另一种方式使其并行(我建议使用流)。我的问题是如何通过打印一些输出来加快线程速度?或者,如果您更喜欢为什么速度太慢?
我尝试改变线程数:一个线程(!)消失了怪异现象,但出现了2个或更多线程(我测试了我的机器上最多两倍的内核)。
我尝试更改正在处理的列表的大小,但仅证明较小的列表可以更快地处理!
我尝试改变输出:两个println输出似乎将速度提高了一个以上,并且在线程寿命结束时进行的打印效果比开始时要差。
我确保它不是GC效果(巨大的堆,并检查了GC日志)。
这是我正在运行的代码。
import java.util.ArrayList;
import java.util.Iterator;
import java.util.LinkedList;
import java.util.List;
public class ListAddThreads {
private static final int SIZE = 20000000;
private static final int REPEATS = 3;
private static final int THREADS = Runtime.getRuntime().availableProcessors() / 2;
private static String log = "none";
public static void main( String[] args ) {
if( args.length > 0 ) switch( args[0] ) {
case "some":
case "more":
case "lots":
log = args[0];
}
System.out.println( "Array List:" );
List<Integer>listA = new ArrayList<>(SIZE);
for( int i = 0; i < SIZE; i++ )
listA.add( i );
for( int i = 0; i < REPEATS; i++ )
add( listA );
System.out.println( "\nLinked List:" );
List<Integer>listL = new LinkedList<>(listA);
for( int i = 0; i < REPEATS; i++ )
add( listL );
}
private static class AdderThread extends Thread {
private long result = 0;
private List<Integer> list;
private int from;
private int to;
AdderThread( List<Integer> list, int from, int to ) {
this.list = list;
this.from = from;
this.to = to;
}
long getResult() {
return result;
}
public void run() {
switch( log ) {
case "some":
System.err.println( "string literal" );
break;
case "more":
System.err.println( Thread.currentThread().getName() + " starting" );
break;
case "lots":
System.err.println( "string literal" );
System.err.println( Thread.currentThread().getName() + " starting" );
break;
}
Iterator<Integer> it = list.iterator();
for( int i = 0; i < to; i++ ) {
Integer j = it.next();
if( i >= from )
result += j;
}
}
}
private static void add( List<Integer> list ) {
long start = System.currentTimeMillis();
AdderThread[] adders = new AdderThread[THREADS];
int size = list.size() / THREADS;
for( int i = 0; i < THREADS; i++ ) {
int from = i * size;
int to = i == THREADS - 1 ? list.size() : (i + 1) * size;
adders[i] = new AdderThread( list, from, to );
}
for( int i = 0; i < THREADS; i++ ) {
adders[i].start();
}
long result = 0;
try {
for( int i = 0; i < THREADS; i++ ) {
adders[i].join();
result += adders[i].getResult();
}
} catch ( InterruptedException x ) {
throw new RuntimeException( "Adder thread interrupted.", x );
}
System.out.println( "Sum: " + result + ", time: " + (System.currentTimeMillis() - start) );
}
}
我这样称呼它:
D:\weird> for %G in (none some more lots) do java -Xms4G -Xmx4G -XX:+PrintGCDetails -Xloggc:ListAddThreads_%G_gc.log ListAddThreads %G 2>nul
以下是一些示例输出,该示例显示LinkedList
的时间随着更多输出(被重定向到遗忘)而变得更快:
java -Xms4G -Xmx4G -XX:+PrintGCDetails -Xloggc:ListAddThreads_none_gc.log ListAddThreads none 2>nul
Array List:
Sum: 199999990000000, time: 62
Sum: 199999990000000, time: 157
Sum: 199999990000000, time: 62
Linked List:
Sum: 199999990000000, time: 3343
Sum: 199999990000000, time: 1814
Sum: 199999990000000, time: 1796
java -Xms4G -Xmx4G -XX:+PrintGCDetails -Xloggc:ListAddThreads_some_gc.log ListAddThreads some 2>nul
Array List:
Sum: 199999990000000, time: 50
Sum: 199999990000000, time: 58
Sum: 199999990000000, time: 56
Linked List:
Sum: 199999990000000, time: 2777
Sum: 199999990000000, time: 560
Sum: 199999990000000, time: 527
java -Xms4G -Xmx4G -XX:+PrintGCDetails -Xloggc:ListAddThreads_more_gc.log ListAddThreads more 2>nul
Array List:
Sum: 199999990000000, time: 48
Sum: 199999990000000, time: 56
Sum: 199999990000000, time: 39
Linked List:
Sum: 199999990000000, time: 108
Sum: 199999990000000, time: 93
Sum: 199999990000000, time: 94
java -Xms4G -Xmx4G -XX:+PrintGCDetails -Xloggc:ListAddThreads_lots_gc.log ListAddThreads lots 2>nul
Array List:
Sum: 199999990000000, time: 52
Sum: 199999990000000, time: 68
Sum: 199999990000000, time: 39
Linked List:
Sum: 199999990000000, time: 115
Sum: 199999990000000, time: 89
Sum: 199999990000000, time: 91