https://docs.oracle.com/javase/8/docs/platform/serialization/spec/serialTOC.html

默认序列化有点冗长，并且假定序列化对象的使用场景尽可能广泛，因此默认格式(Serializable)用关于序列化对象的类的信息注释结果流。

外部化使对象流的生产者能够完全控制精确的类元数据(如果有的话)，而不仅仅是类所需的最小标识(例如它的名称)。这在某些情况下显然是可取的，比如在封闭环境中，对象流的生产者和消费者(从流中具体化对象)是匹配的，关于类的额外元数据没有任何作用，而且会降低性能。

此外(正如Uri指出的那样)外部化还提供了对与Java类型对应的流中数据编码的完全控制。对于(一个人为的)例子，您可能希望将布尔值true记录为“Y”，将false记录为“N”。外部化可以让你做到这一点。

Serializable和Externalizable之间的主要区别

Marker interface: Serializable is marker interface without any methods. Externalizable interface contains two methods: writeExternal() and readExternal(). Serialization process: Default Serialization process will be kicked-in for classes implementing Serializable interface. Programmer defined Serialization process will be kicked-in for classes implementing Externalizable interface. Maintenance: Incompatible changes may break serialisation. Backward Compatibility and Control: If you have to support multiple versions, you can have full control with Externalizable interface. You can support different versions of your object. If you implement Externalizable, it's your responsibility to serialize super class public No-arg constructor: Serializable uses reflection to construct object and does not require no arg constructor. But Externalizable demands public no-arg constructor.

更多细节请参考Hitesh Garg的博客。

序列化使用某些默认行为来存储对象并稍后重新创建对象。您可以指定以何种顺序或如何处理引用和复杂的数据结构，但最终还是要为每个基本数据字段使用默认行为。

在极少数情况下使用外部化，您确实希望以完全不同的方式存储和重新构建对象，并且不使用数据字段的默认序列化机制。例如，假设您有自己独特的编码和压缩方案。

https://docs.oracle.com/javase/8/docs/platform/serialization/spec/serialTOC.html

默认序列化有点冗长，并且假定序列化对象的使用场景尽可能广泛，因此默认格式(Serializable)用关于序列化对象的类的信息注释结果流。

外部化使对象流的生产者能够完全控制精确的类元数据(如果有的话)，而不仅仅是类所需的最小标识(例如它的名称)。这在某些情况下显然是可取的，比如在封闭环境中，对象流的生产者和消费者(从流中具体化对象)是匹配的，关于类的额外元数据没有任何作用，而且会降低性能。

此外(正如Uri指出的那样)外部化还提供了对与Java类型对应的流中数据编码的完全控制。对于(一个人为的)例子，您可能希望将布尔值true记录为“Y”，将false记录为“N”。外部化可以让你做到这一点。

一些差异:

For Serialization there is no need of default constructor of that class because Object because JVM construct the same with help of Reflection API. In case of Externalization contructor with no arg is required, because the control is in hand of programmar and later on assign the deserialized data to object via setters. In serialization if user want to skip certain properties to be serialized then has to mark that properties as transient, vice versa is not required for Externalization. When backward compatiblity support is expected for any class then it is recommended to go with Externalizable. Serialization supports defaultObject persisting and if object structure is broken then it will cause issue while deserializing.

There are so many difference exist between Serializable and Externalizable but when we compare difference between custom Serializable(overrided writeObject() & readObject()) and Externalizable then we find that custom implementation is tightly bind with ObjectOutputStream class where as in Externalizable case , we ourself provide an implementation of ObjectOutput which may be ObjectOutputStream class or it could be some other like org.apache.mina.filter.codec.serialization.ObjectSerializationOutputStream

如果是Externalizable接口

@Override
public void writeExternal(ObjectOutput out) throws IOException {
    out.writeUTF(key);
    out.writeUTF(value);
    out.writeObject(emp);
}

@Override
public void readExternal(ObjectInput in) throws IOException, ClassNotFoundException {
    this.key = in.readUTF();
    this.value = in.readUTF();
    this.emp = (Employee) in.readObject();
}





**In case of Serializable interface**


        /* 
             We can comment below two method and use default serialization process as well
             Sequence of class attributes in read and write methods MUST BE same.
        // below will not work it will not work . 
        // Exception = java.io.StreamCorruptedException: invalid type code: 00\
              private void writeObject(java.io.ObjectOutput stream) 
              */
            private void writeObject(java.io.ObjectOutputStream Outstream)
                    throws IOException {

                System.out.println("from writeObject()");
                /*     We can define custom validation or business rules inside read/write methods.
 This way our validation methods will be automatically 
    called by JVM, immediately after default serialization 
    and deserialization process 
    happens.
                 checkTestInfo();
                */

                stream.writeUTF(name);
                stream.writeInt(age);
                stream.writeObject(salary);
                stream.writeObject(address);
            }

            private void readObject(java.io.ObjectInputStream Instream)
                    throws IOException, ClassNotFoundException {
                System.out.println("from readObject()");
                name = (String) stream.readUTF();
                age = stream.readInt();
                salary = (BigDecimal) stream.readObject();
                address = (Address) stream.readObject();
                // validateTestInfo();
            }

为了更好地解释，我添加了示例代码。请检查Externalizable的入/出对象情况。它们不直接绑定到任何实现。 其中Outstream/ stream与类紧密绑定。我们可以扩展ObjectOutputStream/ObjectInputStream，但使用起来有点困难。