Completable Futures

Repository : https://github.com/jorgetovar/completable_futures

Do you know that parallel stream in Java has performance limitations due to Java will use the number of available processors?

If you want to take advance of the latest Java features and overcome this problem, use CompletableFuture.

You can create a ThreadPool greater than the number of available processors and reap the performance improvements. CompletableFuture uses lambdas, and the code is expressive.

My machine has 8 processors

   @Test
    public void numberOfProcessors() {
        assertThat(availableProcessors, is(equalTo(8)));
    }

1 Second: CompletableFuture implementation has the best performance

    @Test
    public void completableFuture_whenBooksAreMoreThanNumberOfProcessors() {
        Executor executor = Executors.newFixedThreadPool(10);

        long start = System.currentTimeMillis();
        var futureCategories = getBooks()
                .map(e -> CompletableFuture.supplyAsync(() -> BookClassifier.apply(e), executor))
                .toList();

        var categories = futureCategories.stream()
                .map(CompletableFuture::join).toList();
        int timeInSeconds = getTimeInSeconds(start);
        assertThat(categories.size(), is(equalTo(10)));
        assertThat(timeInSeconds, OrderingComparison.greaterThanOrEqualTo(1));
        assertThat(timeInSeconds, OrderingComparison.lessThanOrEqualTo(1));

    }

2 Seconds: Parallel Stream implementation has some limitations

Because our current limitation would have to process 8 books first and then 2 more

    @Test
public void parallelStream_whenBooksAreMoreThanNumberOfProcessors() {
        long start = System.currentTimeMillis();
        var categories = getBooks()
        .parallel()
        .map(BookClassifier::apply)
        .toList();
        int timeInSeconds = getTimeInSeconds(start);

        assertThat(categories.size(), is(equalTo(10)));
        assertThat(timeInSeconds, OrderingComparison.greaterThanOrEqualTo(2));
        assertThat(timeInSeconds, OrderingComparison.lessThanOrEqualTo(2));
        }

1 Second: Parallel Stream implementation within the limitations

Because we are within the limit of the available processors


    @Test
    public void parallelStream_whenBooksAreLessThanNumberOfProcessors() {
        int limit = availableProcessors - 1;
        long start = System.currentTimeMillis();
        var categories = getBooks()
                .limit(limit)
                .parallel()
                .map(BookClassifier::apply)
                .toList();
        int timeInSeconds = getTimeInSeconds(start);

        System.out.printf("The operation took %s ms%n", timeInSeconds - start);
        assertThat(categories.size(), is(equalTo(limit)));
        assertThat(timeInSeconds, OrderingComparison.greaterThanOrEqualTo(1));
        assertThat(timeInSeconds, OrderingComparison.lessThanOrEqualTo(1));

    }

10 Seconds: Stream Synchronous implementation has the worst performance

   @Test
   public void stream_whenBooksAreLessThanNumberOfProcessors() {
        long start = System.currentTimeMillis();
        var categories = getBooks()
        .map(BookClassifier::apply).toList();

        int timeInSeconds = getTimeInSeconds(start);
        assertThat(categories.size(), is(equalTo(10)));
        assertThat(timeInSeconds, OrderingComparison.greaterThanOrEqualTo(9));
        assertThat(timeInSeconds, OrderingComparison.lessThanOrEqualTo(10));
        System.out.printf("The stream operation took %s ms%n", timeInSeconds);

        }