kalor

导航

 

序列化是将对象转化为字节流的方法,序列化目的有:

1> 进程间通信;

2> 数据持久性存储。

RPC(Remote Procedure Call Protocol)——远程过程调用协议,它是一种通过网络从远程计算机程序上请求服务,而不需要了解底层网络技术的协议。RPC协议假定某些传输协议的存在,如TCP或UDP,为通信程序之间携带信息数据。在OSI网络通信模型中,RPC跨越了传输层和应用层。RPC使得开发包括网络分布式多程序在内的应用程序更加容易。

Hadoop采用RPC来实现进程间的通信。Generally,RPC的序列化机制有以下特点:

1> 紧凑:紧凑的格式可以利用带宽,加快传输速度;

2> 快速:能减少序列化和反序列化的开销,这会有效减少进程间通信的时间;

3> 可扩展:可以逐步改变,是Client与Server端直接相关的。例如,可以随时加入一个新的参数方法调用;

4> 互操作性:支持不同语言编写的Client和Server端交换Data。

在Hadoop中,序列化处于核心地位。因为无论是存储文件还是在计算中传输数据,都需要执行序列化的过程。序列化与反序列化的速度,序列化后的data大小等都会影响数据传输的速度,以致影响计算的效率。Hadoop并没有采用Java的序列化机制,而是重新写了一个序列化机制Writable(具有紧凑、快速但不易扩展,亦不利于不同语言的互操作),并允许对自己定义的类加入序列化与反序列化方法. 当要在进程间传递对象或持久化对象的时候,就需要序列化对象成字节流,反之当要将接收到或从磁盘读取的字节流转换为对象,就要进行反序列化。Writable是Hadoop的序列化格式,Hadoop定义了这样一个Writable接口。

public interface Writable {  

   // Serialize the fields of this object to out.

   // @param(out) DataOuput (to serialize this object into). @throws IOException    

  void write(DataOutput out) throws IOException;

  // Deserialize the fields of this object from in. For efficiency, implementations should attempt to re-use storage in the existing object where possible.

  // @param(in) DataInput (to deseriablize this object from). @throws IOException

  void readFields(DataInput in) throws IOException;

}

Writable是Hadoop的核心,Hadoop通过它定义了Hadoop中基本的数据类型及其操作。Generally,无论是上传下载data还是运行MapReduce程序,都需使用Writable类。

//WritableComparable can be compared to each other, typically via Comparator. Any type which is to be used as a key in the Hadoop Map-Reduce framework should implement this interface.

public interface WritableComparable<T> extends Writable, Comparable<T> { }

看看一个WritableComparable的具体实例:

/** A WritableComparable for ints. */

public class IntWritable implements WritableComparable {

      private int value;

      public IntWritable() {}

      public IntWritable(int value) { set(value); }

      /** Set the value of this IntWritable. */  

      public void set(int value) { this.value = value; }

      /** Return the value of this IntWritable. */  

      public int get() { return value; }

      public void readFields(DataInput in) throws IOException {     value = in.readInt();   }

      public void write(DataOutput out) throws IOException {     out.writeInt(value);   }

      /** Returns true if o is a IntWritable with the same value. */  

      public boolean equals(Object o) {   

           if (!(o instanceof IntWritable))      

           {  return false; }  

           IntWritable other = (IntWritable)o;    

           return this.value == other.value;  

      }

      public int hashCode() {     return value;   }

      /** Compares two IntWritables. */  

      public int compareTo(Object o) {    

           int thisValue = this.value;    

           int thatValue = ((IntWritable)o).value;    

           return (thisValue<thatValue ? -1 : (thisValue==thatValue ? 0 : 1));  

      }

      public String toString() {     return Integer.toString(value);   }

      /** A Comparator optimized for IntWritable. */

      public static class Comparator extends WritableComparator {    

           public Comparator() {    super(IntWritable.class);    }

           public int compare(byte[] b1, int s1, int l1, byte[] b2, int s2, int l2) {      

                int thisValue = readInt(b1, s1);     

                int thatValue = readInt(b2, s2);     

                return (thisValue<thatValue ? -1 : (thisValue==thatValue ? 0 : 1));    

           }  

       }  

       // register this comparator       

       static {  WritableComparator.define(IntWritable.class, new Comparator());  }

}

代码中的static块调用WritableComparator的static方法define()用来注册上面这个Comparator,就是将其加入WritableComparator的comparators成员中,comparators是HashMap类型且是static的。这样,就告诉WritableComparator,当我使用WritableComparator.get(IntWritable.class)方法的时候,你返回我注册的这个Comparator给我【对IntWritable来说就是IntWritable.Comparator】,然后我就可以使用comparator.compare(byte[] b1, int s1, int l1,byte[] b2, int s2, int l2)来比较b1和b2而不需要将它们反序列化成对象。comparator.compare(byte[] b1, int s1, int l1,byte[] b2, int s2, int l2)中的readInt()是从WritableComparator继承来的,它将IntWritable的value从byte数组中通过移位转换出来。

相关调用如下:

//params byte[] b1, byte[] b2  
RawComparator<IntWritable>comparator = WritableComparator.get(IntWritable.class);  
comparator.compare(b1,0,b1.length,b2,0,b2.length);  

 

注意,当comparators中没有注册要比较的类的Comparator,则会返回一个默认的Comparator,然后使用这个默认Comparator的compare(byte[] b1, int s1, int l1,byte[] b2, int s2, int l2)方法比较b1、b2的时候还是要序列化成对象的,详见后面细讲WritableComparator。

另外关于WritableComparator类定义如下(上面用到过):

 1 public class WritableComparator implements RawComparator {
 2 
 3   private static HashMap<Class, WritableComparator> comparators =
 4     new HashMap<Class, WritableComparator>(); // registry
 5 
 6   /** Get a comparator for a {@link WritableComparable} implementation. */
 7   public static synchronized WritableComparator get(Class<? extends WritableComparable> c) {
 8     WritableComparator comparator = comparators.get(c);
 9     if (comparator == null)
10       comparator = new WritableComparator(c, true);
11     return comparator;
12   }
13 
14   /** Register an optimized comparator for a {@link WritableComparable}
15    * implementation. */
16   public static synchronized void define(Class c,
17                                          WritableComparator comparator) {
18     comparators.put(c, comparator);
19   }
20   .......
21 }

 

 

 

 

 

 

 

posted on 2013-01-06 19:43  kalor  阅读(312)  评论(0编辑  收藏  举报