Java8 Lambda 性能与公共(public)函数

Question

我一直在使用 Java8 VS 对 lambda 性能进行一些演示测试。 Java8 公共(public)函数。

案例如下:

1. 我有一个 10 人的名单(5 男 5 女)。

2. 我想知道哪个女人的年龄在 18 到 25 岁之间

现在，当我执行这些步骤一百万次时，结果将是:

Lambda with ForEach took: 395 ms (396 ms using JUnit)

Public functions took: 173 ms (169 ms using JUnit)

Lambda with Collect took: 334 ms (335 ms using JUnit)

现在我没想到 lambda 的执行时间比常规函数长两倍到六倍。

所以，现在我很想知道我是否在这里遗漏了什么。

可以在这里找到源代码:pastebin.com/BJBk4Tu6

---

跟进:

1. 将列表扩展到 1.000.000 项时
2. 并对所有年轻成年女性进行一次过滤

结果是:

Lambda with ForEach took: 59 ms

Public functions took: 15 ms

Lambda with Collect took: 12 ms

但是，当我尝试过滤同一个包含 1.000.000 人的列表 100 次时，结果将是:

Lambda with ForEach took: 227 ms

Public functions took: 134 ms

Lambda with Collect took: 172 ms

因此，作为最终结论:Lambda 在过滤较大列表时更快，而公共(public)函数(旧方法)在过滤较小列表时更快。

此外，无论您出于何种目的需要多次过滤任何列表，公共(public)函数都会更快。

最新代码:pastebin.com/LcVhgnYv

Answer 1

正如评论中所指出的:您很难从这样一个单一、简单和孤立的微基准测试运行中得出任何结论。

部分引用自another (otherwise unrelated) answer :

In order to properly and reliably measure execution times, there exist several options. Apart from a profiler, like VisualVM, there are frameworks like JMH or Caliper, but admittedly, using them may be some effort.

For the simplest form of a very basic, manual Java Microbenchmark you have to consider the following:

• Run the algorithms multiple times, to give the JIT a chance to kick in
• Run the algorithms alternatingly and not only one after the other
• Run the algorithms with increasing input size
• Somehow save and print the results of the computation, to prevent the computation from being optimized away
• Consider that timings may be distorted by the garbage collector (GC)

These are only rules of thumb, and there may still be unexpected results (refer to the links above for more details). But with this strategy, you usually obtain a good indication about the performance, and at least can see whether it's likely that there really are significant differences between the algorithms.

Related reading:

• How do I write a correct micro-benchmark in Java?
• Java theory and practice: Anatomy of a flawed microbenchmark
• HotSpot Internals

我将这些基本步骤应用于您的程序。这是一个 MCVE :

NOTE: The remaining part was updated in response to the follow-up edit of the question)

import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import java.util.Random;
import java.util.stream.Collectors;

class Person {
    public static final int MALE = 0;
    public static final int FEMALE = 1;
    private final String name;
    private final int sex;
    private final int age;

    public Person(String name, int sex, int age) {
        this.name = name;
        this.sex = sex;
        this.age = age;
    }

    public int getSex() {
        return sex;
    }

    public int getAge() {
        return age;
    }
}

public class Main {

    public static void main(String[] args) {
        new Main();
    }

    private List<Person> people;

    public Main() {

        for (int size=10; size<=1000000; size*=10) {

            Random r = new Random(0);
            people = new ArrayList<Person>();
            for (int i = 0; i < size; i++) {
                int s = r.nextInt(2);
                int a = 25 + r.nextInt(20);
                people.add(new Person("p" + i, s, a));
            }

            int min = 10000000 / size;
            int max = 10 * min;
            for (int n = min; n <= max; n += min) {
                lambdaMethodUsingForEach(n);
                lambdaMethodUsingCollect(n);
                defaultMethod(n);
            }
        }
    }

    public void lambdaMethodUsingForEach(int n) {
        List<Person> lambdaOutput = new ArrayList<Person>();
        long lambdaStart = System.currentTimeMillis();
        for (int i = 0; i < n; i++) {
            lambdaOutput.addAll(getFemaleYoungAdultsUsingLambdaUsingForEach());
        }
        System.out.printf("List size: %10d, runs: %10d, result: %10d, ForEach took: " +
            (System.currentTimeMillis() - lambdaStart) + " ms\n",
            people.size(), n, lambdaOutput.size());
    }

    public void lambdaMethodUsingCollect(int n) {
        List<Person> lambdaOutput = new ArrayList<Person>();
        long lambdaStart = System.currentTimeMillis();
        for (int i = 0; i < n; i++) {
            lambdaOutput.addAll(getFemaleYoungAdultsUsingLambdaUsingCollect());
        }
        System.out.printf("List size: %10d, runs: %10d, result: %10d, collect took: " +
            (System.currentTimeMillis() - lambdaStart) + " ms\n",
            people.size(), n, lambdaOutput.size());
    }

    public void defaultMethod(int n) {
        List<Person> defaultOutput = new ArrayList<Person>();
        long defaultStart = System.currentTimeMillis();
        for (int i = 0; i < n; i++) {
            defaultOutput.addAll(getFemaleYoungAdultsUsingFunctions());
        }
        System.out.printf("List size: %10d, runs: %10d, result: %10d, default took: " +
            (System.currentTimeMillis() - defaultStart) + " ms\n",
            people.size(), n, defaultOutput.size());
    }

    public List<Person> getFemaleYoungAdultsUsingLambdaUsingForEach() {
        List<Person> people = new ArrayList<Person>();
        this.people.stream().filter(
                (p) -> p.getSex() == Person.FEMALE &&
                p.getAge() >= 18 &&
                p.getAge() <= 25).forEach(people::add);
        return people;
    }

    public List<Person> getFemaleYoungAdultsUsingLambdaUsingCollect() {
        return this.people.stream().filter(
                (p) -> p.getSex() == Person.FEMALE &&
                p.getAge() >= 18 &&
                p.getAge() <= 25).collect(Collectors.toList());
    }

    public List<Person> getFemaleYoungAdultsUsingFunctions() {
        List<Person> people = new ArrayList<Person>();
        for (Person p : this.people) {
            if (p.getSex() == Person.FEMALE && p.getAge() >= 18 && p.getAge() <= 25) {
                people.add(p);
            }
        }
        return people;
    }
}

My Machine® 上的输出是这样的:

    ...
List size:       10, runs:   10000000, result:   10000000, ForEach took: 1482 ms
List size:       10, runs:   10000000, result:   10000000, collect took: 2014 ms
List size:       10, runs:   10000000, result:   10000000, default took: 1013 ms
...
List size:      100, runs:    1000000, result:    3000000, ForEach took: 664 ms
List size:      100, runs:    1000000, result:    3000000, collect took: 515 ms
List size:      100, runs:    1000000, result:    3000000, default took: 441 ms
...
List size:     1000, runs:     100000, result:    2300000, ForEach took: 778 ms
List size:     1000, runs:     100000, result:    2300000, collect took: 721 ms
List size:     1000, runs:     100000, result:    2300000, default took: 841 ms
...
List size:    10000, runs:      10000, result:    2450000, ForEach took: 970 ms
List size:    10000, runs:      10000, result:    2450000, collect took: 971 ms
List size:    10000, runs:      10000, result:    2450000, default took: 1119 ms
...
List size:   100000, runs:       1000, result:    2536000, ForEach took: 976 ms
List size:   100000, runs:       1000, result:    2536000, collect took: 1057 ms
List size:   100000, runs:       1000, result:    2536000, default took: 1109 ms
...
List size:  1000000, runs:        100, result:    2488600, ForEach took: 1323 ms
List size:  1000000, runs:        100, result:    2488600, collect took: 1305 ms
List size:  1000000, runs:        100, result:    2488600, default took: 1422 ms

您可以看到 ForEach 和 default(公共(public)方法)方法之间的区别正在消失，即使对于较小的列表也是如此。对于更大的列表，基于 lambda 的方法似乎甚至有一点优势。

再次强调:这是一个非常简单的微基准测试，即使这样也不一定能说明这些方法在实践中的性能。但是，至少可以合理地假设 ForEach 和公共(public)方法之间的差异没有初始测试建议的那么大。 Nevertleless:我为在 JMH 或 Caliper 中运行此程序并发布了一些关于此的进一步见解的任何人 +1。