如何合理地估算线程池大小

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：

1Speedup <= 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：

001package pool_size_calculate;

002 

003import java.math.BigDecimal;

004import java.math.RoundingMode;

005import java.util.Timer;

006import java.util.TimerTask;

007import 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 */

017public 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 voidcalculateBoundaries(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：

01package pool_size_calculate;

02 

03import java.io.BufferedReader;

04import java.io.IOException;

05import java.io.InputStreamReader;

06import java.lang.management.ManagementFactory;

07import java.math.BigDecimal;

08import java.net.HttpURLConnection;

09import java.net.URL;

10import java.util.concurrent.BlockingQueue;

11import java.util.concurrent.LinkedBlockingQueue;

12 

13public 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), newBigDecimal(100000));

33    }

34 

35}

36 

37/**

38 * 自定义的异步IO任务

39 * @author Will

40 *

41 */

42class 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：

01Target queue memory usage (bytes): 100000

02createTask() produced pool_size_calculate.AsyncIOTask which took 40 bytes in a queue

03Formula: 100000 / 40

04* Recommended queue capacity (bytes): 2500

05Number of CPU: 4

06Target utilization: 1

07Elapsed time (nanos): 3000000000

08Compute time (nanos): 47181000

09Wait time (nanos): 2952819000

10Formula: 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：

1ThreadPoolExecutor pool =

2 new ThreadPoolExecutor(256, 256, 0L, TimeUnit.MILLISECONDS, newLinkedBlockingQueue(2500));

转载自并发编程网 – ifeve.com原文链接地址: 如何合理地估算线程池大小？