如何合理地估算线程池大小？

转载自：http://ifeve.com/how-to-calculate-threadpool-size/

如何合理地估算线程池大小？

这个问题虽然看起来很小，却并不那么容易回答。大家如果有更好的方法欢迎赐教，先来一个天真的估算方法：假设要求一个系统的TPS（Transaction Per Second或者Task Per Second）至少为20，然后假设每个Transaction由一个线程完成，继续假设平均每个线程处理一个Transaction的时间为4s。那么问题转化为：

如何设计线程池大小，使得可以在1s内处理完20个Transaction？

计算过程很简单，每个线程的处理能力为0.25TPS，那么要达到20TPS，显然需要20/0.25=80个线程。

很显然这个估算方法很天真，因为它没有考虑到CPU数目。一般服务器的CPU核数为16或者32，如果有80个线程，那么肯定会带来太多不必要的线程上下文切换开销。

再来第二种简单的但不知是否可行的方法（N为CPU总核数）：

• 如果是CPU密集型应用，则线程池大小设置为N+1
• 如果是IO密集型应用，则线程池大小设置为2N+1

如果一台服务器上只部署这一个应用并且只有这一个线程池，那么这种估算或许合理，具体还需自行测试验证。

接下来在这个文档：服务器性能IO优化 中发现一个估算公式：

1	最佳线程数目 = （（线程等待时间+线程CPU时间）/线程CPU时间 ）* CPU数目

比如平均每个线程CPU运行时间为0.5s，而线程等待时间（非CPU运行时间，比如IO）为1.5s，CPU核心数为8，那么根据上面这个公式估算得到：((0.5+1.5)/0.5)*8=32。这个公式进一步转化为：

1	最佳线程数目 = （线程等待时间与线程CPU时间之比 + 1）* CPU数目

可以得出一个结论：

线程等待时间所占比例越高，需要越多线程。线程CPU时间所占比例越高，需要越少线程。

上一种估算方法也和这个结论相合。

一个系统最快的部分是CPU，所以决定一个系统吞吐量上限的是CPU。增强CPU处理能力，可以提高系统吞吐量上限。但根据短板效应，真实的系统吞吐量并不能单纯根据CPU来计算。那要提高系统吞吐量，就需要从“系统短板”（比如网络延迟、IO）着手：

• 尽量提高短板操作的并行化比率，比如多线程下载技术
• 增强短板能力，比如用NIO替代IO

第一条可以联系到Amdahl定律，这条定律定义了串行系统并行化后的加速比计算公式：

1	加速比=优化前系统耗时 / 优化后系统耗时

加速比越大，表明系统并行化的优化效果越好。Addahl定律还给出了系统并行度、CPU数目和加速比的关系，加速比为Speedup，系统串行化比率（指串行执行代码所占比率）为F，CPU数目为N：

1	Speedup <= 1 / (F + (1-F)/N)

当N足够大时，串行化比率F越小，加速比Speedup越大。

写到这里，我突然冒出一个问题。

是否使用线程池就一定比使用单线程高效呢？

答案是否定的，比如Redis就是单线程的，但它却非常高效，基本操作都能达到十万量级/s。从线程这个角度来看，部分原因在于：

• 多线程带来线程上下文切换开销，单线程就没有这种开销
• 锁

当然“Redis很快”更本质的原因在于：Redis基本都是内存操作，这种情况下单线程可以很高效地利用CPU。而多线程适用场景一般是：存在相当比例的IO和网络操作。

所以即使有上面的简单估算方法，也许看似合理，但实际上也未必合理，都需要结合系统真实情况（比如是IO密集型或者是CPU密集型或者是纯内存操作）和硬件环境（CPU、内存、硬盘读写速度、网络状况等）来不断尝试达到一个符合实际的合理估算值。

最后来一个“Dark Magic”估算方法（因为我暂时还没有搞懂它的原理），使用下面的类：

package pool_size_calculate;
002
 
003
import java.math.BigDecimal;
004
import java.math.RoundingMode;
005
import java.util.Timer;
006
import java.util.TimerTask;
007
import java.util.concurrent.BlockingQueue;
008
 
009
/**
010
 * A class that calculates the optimal thread pool boundaries. It takes the
011
 * desired target utilization and the desired work queue memory consumption as
012
 * input and retuns thread count and work queue capacity.
013
 *
014
 * @author Niklas Schlimm
015
 *
016
 */
017
public abstract class PoolSizeCalculator {
018
 
019
    /**
020
     * The sample queue size to calculate the size of a single {@link Runnable}
021
     * element.
022
     */
023
    private final int SAMPLE_QUEUE_SIZE = 1000;
024
 
025
    /**
026
     * Accuracy of test run. It must finish within 20ms of the testTime
027
     * otherwise we retry the test. This could be configurable.
028
     */
029
    private final int EPSYLON = 20;
030
 
031
    /**
032
     * Control variable for the CPU time investigation.
033
     */
034
    private volatile boolean expired;
035
 
036
    /**
037
     * Time (millis) of the test run in the CPU time calculation.
038
     */
039
    private final long testtime = 3000;
040
 
041
    /**
042
     * Calculates the boundaries of a thread pool for a given {@link Runnable}.
043
     *
044
     * @param targetUtilization
045
     *            the desired utilization of the CPUs (0 <= targetUtilization <=   *            1)     * @param targetQueueSizeBytes   *            the desired maximum work queue size of the thread pool (bytes)     */     protected void calculateBoundaries(BigDecimal targetUtilization,            BigDecimal targetQueueSizeBytes) {      calculateOptimalCapacity(targetQueueSizeBytes);         Runnable task = creatTask();        start(task);        start(task); // warm up phase       long cputime = getCurrentThreadCPUTime();       start(task); // test intervall      cputime = getCurrentThreadCPUTime() - cputime;      long waittime = (testtime * 1000000) - cputime;         calculateOptimalThreadCount(cputime, waittime, targetUtilization);  }   private void calculateOptimalCapacity(BigDecimal targetQueueSizeBytes) {        long mem = calculateMemoryUsage();      BigDecimal queueCapacity = targetQueueSizeBytes.divide(new BigDecimal(              mem), RoundingMode.HALF_UP);        System.out.println("Target queue memory usage (bytes): "                + targetQueueSizeBytes);        System.out.println("createTask() produced "                 + creatTask().getClass().getName() + " which took " + mem               + " bytes in a queue");         System.out.println("Formula: " + targetQueueSizeBytes + " / " + mem);       System.out.println("* Recommended queue capacity (bytes): "                 + queueCapacity);   }   /**      * Brian Goetz' optimal thread count formula, see 'Java Concurrency in   * Practice' (chapter 8.2)   *       * @param cpu    *            cpu time consumed by considered task   * @param wait   *            wait time of considered task   * @param targetUtilization      *            target utilization of the system   */     private void calculateOptimalThreadCount(long cpu, long wait,           BigDecimal targetUtilization) {         BigDecimal waitTime = new BigDecimal(wait);         BigDecimal computeTime = new BigDecimal(cpu);       BigDecimal numberOfCPU = new BigDecimal(Runtime.getRuntime()                .availableProcessors());        BigDecimal optimalthreadcount = numberOfCPU.multiply(targetUtilization)                 .multiply(                      new BigDecimal(1).add(waitTime.divide(computeTime,                              RoundingMode.HALF_UP)));        System.out.println("Number of CPU: " + numberOfCPU);        System.out.println("Target utilization: " + targetUtilization);         System.out.println("Elapsed time (nanos): " + (testtime * 1000000));        System.out.println("Compute time (nanos): " + cpu);         System.out.println("Wait time (nanos): " + wait);       System.out.println("Formula: " + numberOfCPU + " * "                + targetUtilization + " * (1 + " + waitTime + " / "                 + computeTime + ")");       System.out.println("* Optimal thread count: " + optimalthreadcount);    }   /**      * Runs the {@link Runnable} over a period defined in {@link #testtime}.     * Based on Heinz Kabbutz' ideas     * (http://www.javaspecialists.eu/archive/Issue124.html).    *       * @param task   *            the runnable under investigation   */     public void start(Runnable task) {      long start = 0;         int runs = 0;       do {            if (++runs > 5) {
046
                throw new IllegalStateException("Test not accurate");
047
            }
048
            expired = false;
049
            start = System.currentTimeMillis();
050
            Timer timer = new Timer();
051
            timer.schedule(new TimerTask() {
052
                public void run() {
053
                    expired = true;
054
                }
055
            }, testtime);
056
            while (!expired) {
057
                task.run();
058
            }
059
            start = System.currentTimeMillis() - start;
060
            timer.cancel();
061
        } while (Math.abs(start - testtime) > EPSYLON);
062
        collectGarbage(3);
063
    }
064
 
065
    private void collectGarbage(int times) {
066
        for (int i = 0; i < times; i++) {
067
            System.gc();
068
            try {
069
                Thread.sleep(10);
070
            } catch (InterruptedException e) {
071
                Thread.currentThread().interrupt();
072
                break;
073
            }
074
        }
075
    }
076
 
077
    /**
078
     * Calculates the memory usage of a single element in a work queue. Based on
079
     * Heinz Kabbutz' ideas
080
     * (http://www.javaspecialists.eu/archive/Issue029.html).
081
     *
082
     * @return memory usage of a single {@link Runnable} element in the thread
083
     *         pools work queue
084
     */
085
    public long calculateMemoryUsage() {
086
        BlockingQueue queue = createWorkQueue();
087
        for (int i = 0; i < SAMPLE_QUEUE_SIZE; i++) {
088
            queue.add(creatTask());
089
        }
090
        long mem0 = Runtime.getRuntime().totalMemory()
091
                - Runtime.getRuntime().freeMemory();
092
        long mem1 = Runtime.getRuntime().totalMemory()
093
                - Runtime.getRuntime().freeMemory();
094
        queue = null;
095
        collectGarbage(15);
096
        mem0 = Runtime.getRuntime().totalMemory()
097
                - Runtime.getRuntime().freeMemory();
098
        queue = createWorkQueue();
099
        for (int i = 0; i < SAMPLE_QUEUE_SIZE; i++) {
100
            queue.add(creatTask());
101
        }
102
        collectGarbage(15);
103
        mem1 = Runtime.getRuntime().totalMemory()
104
                - Runtime.getRuntime().freeMemory();
105
        return (mem1 - mem0) / SAMPLE_QUEUE_SIZE;
106
    }
107
 
108
    /**
109
     * Create your runnable task here.
110
     *
111
     * @return an instance of your runnable task under investigation
112
     */
113
    protected abstract Runnable creatTask();
114
 
115
    /**
116
     * Return an instance of the queue used in the thread pool.
117
     *
118
     * @return queue instance
119
     */
120
    protected abstract BlockingQueue createWorkQueue();
121
 
122
    /**
123
     * Calculate current cpu time. Various frameworks may be used here,
124
     * depending on the operating system in use. (e.g.
125
     * http://www.hyperic.com/products/sigar). The more accurate the CPU time
126
     * measurement, the more accurate the results for thread count boundaries.
127
     *
128
     * @return current cpu time of current thread
129
     */
130
    protected abstract long getCurrentThreadCPUTime();
131
 
132
}
然后自己继承这个抽象类并实现它的三个抽象方法，比如下面是我写的一个示例（任务是请求网络数据），其中我指定期望CPU利用率为1.0（即100%），任务队列总大小不超过100,000字节：

view sourceprint?
01
package pool_size_calculate;
02
 
03
import java.io.BufferedReader;
04
import java.io.IOException;
05
import java.io.InputStreamReader;
06
import java.lang.management.ManagementFactory;
07
import java.math.BigDecimal;
08
import java.net.HttpURLConnection;
09
import java.net.URL;
10
import java.util.concurrent.BlockingQueue;
11
import java.util.concurrent.LinkedBlockingQueue;
12
 
13
public class SimplePoolSizeCaculatorImpl extends PoolSizeCalculator {
14
 
15
    @Override
16
    protected Runnable creatTask() {
17
        return new AsyncIOTask();
18
    }
19
 
20
    @Override
21
    protected BlockingQueue createWorkQueue() {
22
        return new LinkedBlockingQueue(1000);
23
    }
24
 
25
    @Override
26
    protected long getCurrentThreadCPUTime() {
27
        return ManagementFactory.getThreadMXBean().getCurrentThreadCpuTime();
28
    }
29
 
30
    public static void main(String[] args) {
31
        PoolSizeCalculator poolSizeCalculator = new SimplePoolSizeCaculatorImpl();
32
        poolSizeCalculator.calculateBoundaries(new BigDecimal(1.0), new BigDecimal(100000));
33
    }
34
 
35
}
36
 
37
/**
38
 * 自定义的异步IO任务
39
 * @author Will
40
 *
41
 */
42
class AsyncIOTask implements Runnable {
43
 
44
    @Override
45
    public void run() {
46
        HttpURLConnection connection = null;
47
        BufferedReader reader = null;
48
        try {
49
            String getURL = "http://baidu.com";
50
            URL getUrl = new URL(getURL);
51
 
52
            connection = (HttpURLConnection) getUrl.openConnection();
53
            connection.connect();
54
            reader = new BufferedReader(new InputStreamReader(
55
                    connection.getInputStream()));
56
 
57
            String line;
58
            while ((line = reader.readLine()) != null) {
59
                // empty loop
60
            }
61
        }
62
 
63
        catch (IOException e) {
64
 
65
        } finally {
66
            if(reader != null) {
67
                try {
68
                    reader.close();
69
                }
70
                catch(Exception e) {
71
 
72
                }
73
            }
74
            connection.disconnect();
75
        }
76
 
77
    }
78
 
79
}

得到的输出如下：

Target queue memory usage (bytes): 100000
02
createTask() produced pool_size_calculate.AsyncIOTask which took 40 bytes in a queue
03
Formula: 100000 / 40
04
* Recommended queue capacity (bytes): 2500
05
Number of CPU: 4
06
Target utilization: 1
07
Elapsed time (nanos): 3000000000
08
Compute time (nanos): 47181000
09
Wait time (nanos): 2952819000
10
Formula: 4 * 1 * (1 + 2952819000 / 47181000)
11
* Optimal thread count: 256

推荐的任务队列大小为2500，线程数为256，有点出乎意料之外。我可以如下构造一个线程池：

ThreadPoolExecutor pool =
2
 new ThreadPoolExecutor(256, 256, 0L, TimeUnit.MILLISECONDS, new LinkedBlockingQueue(2500));