BlockingQueue vs PipedOutputStream and PipedInputStream

问题

I want to know Advantages to use BlockingQueue instead of (PipedOutputStream and PipedInputStream)

import java.io.*;
import java.util.concurrent.*;


public class PipedStreamVsBlocking {

  public static void main(String... args) {

    BlockingQueue<Integer> blockingQueue = new LinkedBlockingDeque<>(2);
    ExecutorService executor = Executors.newFixedThreadPool(4);
    Runnable producerTask = () -> {
      try {
        while (true) {
          int value = ThreadLocalRandom.current().nextInt(0, 1000);
          blockingQueue.put(value);
          System.out.println("BlockingQueue.Produced " + value);
          int timeSleeping = ThreadLocalRandom.current().nextInt(500, 1000);
          Thread.sleep(timeSleeping);
        }
      } catch (InterruptedException e) {
        e.printStackTrace();
      }
    };
    Runnable consumerTask = () -> {
      try {
        while (true) {
          int value = blockingQueue.take();
          System.out.println("BlockingQueue.Consume " + value);
          int timeSleeping = ThreadLocalRandom.current().nextInt(500, 1000);
          Thread.sleep(timeSleeping);
        }
      } catch (InterruptedException e) {
        e.printStackTrace();
      }
    };

    PipedOutputStream pipedSrc = new PipedOutputStream();
    PipedInputStream pipedSnk = new PipedInputStream();
    try {
      pipedSnk.connect(pipedSrc);
    } catch (IOException e) {
      e.printStackTrace();
    }

    Runnable runnablePut2 = () -> {
      try {
        ObjectOutputStream oos = new ObjectOutputStream(pipedSrc);
        while (true) {
          int value = ThreadLocalRandom.current().nextInt(0, 1000);
          oos.writeInt(value);
          oos.flush();
          System.out.println("PipedStream.Produced " + value);
          int timeSleeping = ThreadLocalRandom.current().nextInt(500, 1000);
          Thread.sleep(timeSleeping);
        }
      } catch (Exception e) {
        e.printStackTrace();
      }
    };

    Runnable runnableGet2 = () -> {
      try {
        ObjectInputStream ois = new ObjectInputStream(pipedSnk);
        while (true) {
          int value = ois.readInt();
          System.out.println("PipedStream.Consume " + value);
          int timeSleeping = ThreadLocalRandom.current().nextInt(500, 1000);
          Thread.sleep(timeSleeping);
        }
      } catch (Exception e) {
        e.printStackTrace();
      }
    };
    executor.execute(producerTask);
    executor.execute(consumerTask);
    executor.execute(runnablePut2);
    executor.execute(runnableGet2);
    executor.shutdown();
  }

}

The output for this code is:

BlockingQueue.Consume 298
BlockingQueue.Produced 298
PipedStream.Produced 510
PipedStream.Consume 510
BlockingQueue.Produced 536
BlockingQueue.Consume 536
PipedStream.Produced 751
PipedStream.Consume 751
PipedStream.Produced 619
BlockingQueue.Produced 584
BlockingQueue.Consume 584
PipedStream.Consume 619
BlockingQueue.Produced 327
PipedStream.Produced 72
BlockingQueue.Consume 327
PipedStream.Consume 72
BlockingQueue.Produced 823
BlockingQueue.Consume 823
PipedStream.Produced 544
PipedStream.Consume 544
BlockingQueue.Produced 352
BlockingQueue.Consume 352
PipedStream.Produced 134
PipedStream.Consume 134

I think that to use PipedStream (PipedOutputStream and PipedInputStream) have advantages, I know when the data is produced/Processed directly.

May be I wrong, And this recommendation to use BlockingQueue instead of Pipe.

But, your comments/recommendations are not found in the documentation. For this reason, I need to know what I missed.

Why should I use BlockingQueue instead of Piped?

回答1:

Like any Java Collection, a BlockingQueue stores references to objects, so the thread(s) retrieving objects from it receive precisely the same runtime objects, the producing thread(s) put into it.

In contrast, Serialization stores a persistent form into the byte stream, which only works for Serializable objects and will lead to the creation of copies at the receiving end. In some cases, the objects may get replaced by canonical objects afterwards, still, the entire procedure is significantly more expensive than just transferring references.

In your example case, where you transfer int values, the object identity doesn’t matter, but the overhead of boxing, serializing, deserializing, and unboxing Integer instances is even more questionable.

If you didn’t use Serialization, but transferred int values as four byte quantities directly, using PipedOutputStream and PipedInputStream had a point, as it is a good tool for transferring large quantities of primitive data. It also has an intrinsic support for marking the end of data by closing the pipe.

These pipes would also be the right tool for software that ought to be agnostic regarding processes or even the computers running the producer or consumer, i.e. when you want to be able to use the same software when the pipe is actually between processes or even a network connection. That would also justify using Serialization (as JMX connections do).

But unless you’re truly transferring single bytes that retain their meaning when being torn apart, there’s the intrinsic limitation that only one producer can write into a pipe and only one consumer can read the data.

来源：https://stackoverflow.com/questions/57049559/blockingqueue-vs-pipedoutputstream-and-pipedinputstream

标签

java-8

executorservice

blockingqueue