我已经看过并尝试过如何在流中求和的各种不同实现。这是我的代码:

List<Person> persons = new ArrayList<Person>();

for(int i=0; i < 10000000; i++){
    persons.add(new Person("random", 26));
}

Long start = System.currentTimeMillis();
int test = persons.stream().collect(Collectors.summingInt(p -> p.getAge()));
Long end = System.currentTimeMillis();
System.out.println("Sum of ages = " + test + " and it took : " + (end - start) + " ms with collectors");

Long start3 = System.currentTimeMillis();
int test3 = persons.parallelStream().collect(Collectors.summingInt(p -> p.getAge()));
Long end3 = System.currentTimeMillis();
System.out.println("Sum of ages = " + test3 + " and it took : " + (end3 - start3) + " ms with collectors and parallel stream");


Long start2 = System.currentTimeMillis();
int test2 = persons.stream().mapToInt(p -> p.getAge()).sum();
Long end2 = System.currentTimeMillis();
System.out.println("Sum of ages = " + test2 + " and it took : " + (end2 - start2) + " ms with map and sum");

Long start4 = System.currentTimeMillis();
int test4 = persons.parallelStream().mapToInt(p -> p.getAge()).sum();
Long end4 = System.currentTimeMillis();
System.out.println("Sum of ages = " + test4 + " and it took : " + (end4 - start4) + " ms with map and sum and parallel stream");

这给了我以下结果:
Sum of ages = 220000000 and it took : 110 ms with collectors
Sum of ages = 220000000 and it took : 272 ms with collectors and parallel stream
Sum of ages = 220000000 and it took : 137 ms with map and sum
Sum of ages = 220000000 and it took : 134 ms with map and sum and parallel stream

我尝试了几次,每次都给我不同的结果(大多数情况下,最后一个解决方案是最好的),所以我想知道:

1)正确的方法是什么?

2)为什么呢? (与其他解决方案有什么区别?)

最佳答案

在我们获得实际答案之前,您应该了解以下几点:

  • 测试结果的差异可能很大,具体取决于许多因素(例如,您正在运行的计算机)。这是在8核计算机上一次运行的结果:
    Sum of ages = 260000000 and it took : 94 ms with collectors
    Sum of ages = 260000000 and it took : 61 ms with collectors and parallel stream
    Sum of ages = 260000000 and it took : 70 ms with map and sum
    Sum of ages = 260000000 and it took : 94 ms with map and sum and parallel stream
    

    然后在以后的运行中:
    Sum of ages = 260000000 and it took : 68 ms with collectors
    Sum of ages = 260000000 and it took : 67 ms with collectors and parallel stream
    Sum of ages = 260000000 and it took : 66 ms with map and sum
    Sum of ages = 260000000 and it took : 109 ms with map and sum and parallel stream
    
  • Micro基准测试不是一个容易的话题。有一些方法可以做到这一点(稍后再介绍),但是在大多数情况下,仅尝试使用System.currentTimeMillies()将无法可靠地工作。
  • 仅仅因为Java 8使并行操作变得容易,并不意味着应该在所有地方都使用它们。并行操作在某些情况下有意义,而在其他情况下则没有意义。

  • 好的,现在让我们看看您使用的各种方法。
  • 顺序收集器:您使用的summingInt收集器具有以下实现:
    public static <T> Collector<T, ?, Integer> summingInt(ToIntFunction<? super T> mapper) {
        return new CollectorImpl<>(
                () -> new int[1],
                (a, t) -> { a[0] += mapper.applyAsInt(t); },
                (a, b) -> { a[0] += b[0]; return a; },
                a -> a[0], Collections.emptySet());
    }
    

    因此,首先将创建一个具有一个元素的新数组。然后,对于流中的每个Person元素,collect函数将使用Person#getAge()函数将年龄作为Integer(而不是int!)检索,并将该年龄添加到先前的年龄(在1D数组中)。最后,在处理完所有流之后,它将从该数组中提取值并将其返回。因此,这里有很多自动装箱和-unboxing操作。
  • 并行收集器:这使用ReferencePipeline#forEach(Consumer)函数来累积从映射函数获得的年龄。再次有很多自动装箱和-unboxing。
  • 顺序图和总和:在这里,您将Stream<Person>映射到IntStream。这意味着一件事是,不再需要自动装箱或-unboxing;在某些情况下,这可以节省大量时间。然后使用以下实现对结果流求和:
    @Override
    public final int sum() {
        return reduce(0, Integer::sum);
    }
    

    此处的reduce函数将调用ReduceOps#ReduceOp#evaluateSequential(PipelineHelper<T> helper, Spliterator<P_IN> spliterator)
    本质上,这将对所有数字使用Integer::sum函数,从0和第一个数字开始,然后从第二个数字开始,以此类推。
  • 并行映射和求和:这里的事情变得很有趣。它使用相同的sum()函数,但是在这种情况下,reduce将调用ReduceOps#ReduceOp#evaluateParallel(PipelineHelper<T> helper, Spliterator<P_IN> spliterator)而不是顺序选项。基本上,这将使用分而治之的方法将这些值相加。现在,分而治之的最大优势当然是可以轻松并行完成。但是,它确实需要多次拆分和重新加入流,这会浪费时间。因此,取决于与元素相关的实际任务的复杂性,它的运行速度可能会有很大差异。在添加的情况下,在大多数情况下可能不值得。从我的结果可以看出,它始终是较慢的方法之一。

  • 现在,要真正了解需要花多长时间,让我们做一个适当的微型基准测试。我将使用带有以下基准代码的JMH:
    package com.stackoverflow.user2352924;
    
    import org.openjdk.jmh.annotations.*;
    
    import java.util.ArrayList;
    import java.util.List;
    import java.util.concurrent.TimeUnit;
    import java.util.stream.Collectors;
    
    @BenchmarkMode(Mode.Throughput)
    @OutputTimeUnit(TimeUnit.MINUTES)
    @Warmup(iterations = 5, time = 5, timeUnit = TimeUnit.SECONDS)
    @Measurement(iterations = 10, time = 10, timeUnit = TimeUnit.SECONDS)
    @State(Scope.Benchmark)
    @Fork(1)
    @Threads(2)
    public class MicroBenchmark {
    
        private static List<Person> persons = new ArrayList<>();
    
        private int test;
    
        static {
            for(int i=0; i < 10000000; i++){
                persons.add(new Person("random", 26));
            }
        }
    
        @Benchmark
        public void sequentialCollectors() {
            test = 0;
            test += persons.stream().collect(Collectors.summingInt(p -> p.getAge()));
        }
    
        @Benchmark
        public void parallelCollectors() {
            test = 0;
            test += persons.parallelStream().collect(Collectors.summingInt(p -> p.getAge()));
        }
    
        @Benchmark
        public void sequentialMapSum() {
            test = 0;
            test += persons.stream().mapToInt(p -> p.getAge()).sum();
        }
    
        @Benchmark
        public void parallelMapSum() {
            test = 0;
            test += persons.parallelStream().mapToInt(p -> p.getAge()).sum();
        }
    
    }
    

    这个Maven项目的pom.xml看起来像这样:
    <project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
             xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
        <modelVersion>4.0.0</modelVersion>
    
        <groupId>com.stackoverflow.user2352924</groupId>
        <artifactId>StackOverflow</artifactId>
        <version>1.0</version>
        <packaging>jar</packaging>
    
        <name>Auto-generated JMH benchmark</name>
    
        <prerequisites>
            <maven>3.0</maven>
        </prerequisites>
    
        <dependencies>
            <dependency>
                <groupId>org.openjdk.jmh</groupId>
                <artifactId>jmh-core</artifactId>
                <version>${jmh.version}</version>
            </dependency>
            <dependency>
                <groupId>org.openjdk.jmh</groupId>
                <artifactId>jmh-generator-annprocess</artifactId>
                <version>${jmh.version}</version>
                <scope>provided</scope>
            </dependency>
        </dependencies>
    
        <properties>
            <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
            <jmh.version>0.9.5</jmh.version>
            <javac.target>1.8</javac.target>
            <uberjar.name>benchmarks</uberjar.name>
        </properties>
    
        <build>
            <plugins>
                <plugin>
                    <groupId>org.apache.maven.plugins</groupId>
                    <artifactId>maven-compiler-plugin</artifactId>
                    <version>3.1</version>
                    <configuration>
                        <compilerVersion>${javac.target}</compilerVersion>
                        <source>${javac.target}</source>
                        <target>${javac.target}</target>
                    </configuration>
                </plugin>
                <plugin>
                    <groupId>org.apache.maven.plugins</groupId>
                    <artifactId>maven-shade-plugin</artifactId>
                    <version>2.2</version>
                    <executions>
                        <execution>
                            <phase>package</phase>
                            <goals>
                                <goal>shade</goal>
                            </goals>
                            <configuration>
                                <finalName>microbenchmarks</finalName>
                                <transformers>
                                    <transformer implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer">
                                        <mainClass>org.openjdk.jmh.Main</mainClass>
                                    </transformer>
                                </transformers>
                            </configuration>
                        </execution>
                    </executions>
                </plugin>
            </plugins>
            <pluginManagement>
                <plugins>
                    <plugin>
                        <artifactId>maven-clean-plugin</artifactId>
                        <version>2.5</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-deploy-plugin</artifactId>
                        <version>2.8.1</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-install-plugin</artifactId>
                        <version>2.5.1</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-jar-plugin</artifactId>
                        <version>2.4</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-javadoc-plugin</artifactId>
                        <version>2.9.1</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-resources-plugin</artifactId>
                        <version>2.6</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-site-plugin</artifactId>
                        <version>3.3</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-source-plugin</artifactId>
                        <version>2.2.1</version>
                    </plugin>
                    <plugin>
                        <artifactId>maven-surefire-plugin</artifactId>
                        <version>2.17</version>
                    </plugin>
                </plugins>
            </pluginManagement>
        </build>
    
    </project>
    

    确保Maven也与Java 8一起运行,否则您将收到难看的错误。

    在这里我不会详细介绍如何使用JMH(在其他地方也可以使用),但这是我得到的结果:
    # Run complete. Total time: 00:08:48
    
    Benchmark                                     Mode  Samples     Score  Score error    Units
    c.s.u.MicroBenchmark.parallelCollectors      thrpt       10  3658,949      775,115  ops/min
    c.s.u.MicroBenchmark.parallelMapSum          thrpt       10  2616,905      221,109  ops/min
    c.s.u.MicroBenchmark.sequentialCollectors    thrpt       10  5502,160      439,024  ops/min
    c.s.u.MicroBenchmark.sequentialMapSum        thrpt       10  6120,162      609,232  ops/min
    

    因此,在我运行这些测试时,在我的系统上,顺序映射总和要快得多,在并行映射总和(使用分治法)仅可以完成2600多次的同时,可以完成6100多次操作实际上,顺序方法都比并行方法快得多。

    现在,在可以更轻松地并行运行的情况下-例如Person#getAge()函数比 getter 要复杂得多-并行方法可能是更好的解决方案。最后,这完全取决于被测试案例中并行运行的效率。

    要记住的另一件事:如果有疑问,请执行适当的微观基准测试。 ;-)

    10-04 12:20