我正在本地创建一个序列化树Map，并将其传输到hadoop2.7.3上的hdfs目录，如下所示。

// Create tree map and seralize to local folder
  String localPath = "/home/test/file"
  FileOutputStream fos = new FileOutputStream(localPath);
  ObjectOutputStream oos = new ObjectOutputStream(fos);
  oos.writeObject(treemap);
  oos.close();
  fos.close();

  // Transfer serialzied file to HDFS 
  Path hdfsPath = new Path("hdfs://localhost:54310/user/filename");
  FileSystem fs = path.getFileSystem(conf);
  InputStream in = new BufferedInputStream(new FileInputStream(localPath));
  OutputStream out = fs.create(hdfsPath);
  IOUtils.copyBytes(in, out, 4096, true);

  // delete local copy
  File f = new File(localPath);
  f.delete();

序列化文件的大小为17 mb。然后我重新读取序列化文件并在稍后反序列化它。

String filename = "hdfs://localhost:54310/user/filename"
 FileSystem fs = FileSystem.get(uri, conf);
 InputStream in = fs.open(new Path(filename));
 ObjectInputStream objReader = new ObjectInputStream(in);
 map = (TreeMap) objReader.readObject();

此反序列化操作本身需要40秒，而程序的其余部分只需4-6秒即可运行。我试着在本地而不是在hdfs文件上运行反序列化，它在几秒钟内就完成了。你知道为什么hdfs要花这么长时间吗？