A General-Purpose Object Marshalling Framework

A General-Purpose Object
Marshalling Framework

David M. Lloyd
Senior Software Engineer
JBoss, a division of Red Hat, Inc.
[email protected]

About the Presenter

Developing software for around 12 years
• Primary focus on networking software (clients and servers)

Started at JBoss in August 2006

Joined the JBoss Remoting team in 2007

• Work was begun on the next-generation Remoting implementation
• This work lead to the development of JBoss Marshalling

What is Marshalling?

In this context, Marshalling is the process of converting
an object or object graph into a stream of bytes.

• “Serialization” and “Marshalling” are synonymous
• JBoss Marshalling is so named to differentiate from standard Java

Serialization, as well as the existing (but different) JBoss
Serialization project

Why isn't Java Serialization good enough?

Monolithic (sometimes confusing) API
• Very large and complex base classes
• Protocol extensions can only be implemented by subclassing the

Object*Stream classes, sometimes with great difficulty
 Customizing class descriptor format

• Some things are almost impossible to change (at least, without resorting to

reflection)
 Customizing (or removing) the stream header format
 Pre-sizing internal tables

Performance limited implementation
• Cannot specify tuning parameters
• Protocol structures are quite large; often unnecessarily so

Why isn't Java Serialization good enough?
(cont.)

Heavyweight construction
• Makes pooling very difficult
• Stream header is written on construction
• Relatively expensive permission check is required to construct an instance

Bugs tied to JDK
• If you find a problem, the only solution is file a bug and wait for the next JDK

release, which might entail re-testing more than just the module which
encountered the problem

Why is JBoss Marshalling better?

Lightweight marshaller/unmarshaller instance creation
• Permission checking is moved to factory construction rather than instance

construction

• Instances may be easily reused if desired

Easy to configure

Ability to support multiple implementations easily
• Marshalling API elements are mostly interfaces

Highly tunable

Very customizable with special marshalling strategies

Constructing Marshallers and Unmarshallers

Every protocol implementation has a corresponding
MarshallerFactory

A MarshallerFactory has a create method for Marshallers
and one for Unmarshallers

A MarshallerConfiguration instance is required by either
method

Usage example: HelloWorld

(example)

Configurable Parameters

Expected instance count
• Number of distinct objects expected to be sent

Expected class count
• Number of distinct classes expected to be sent

Buffer size
• Used by any implementation which employs an internal buffer

Pluggable Features

Stream Header
Class Table
Object Table
Class Resolver
Object Resolver
Class Externalizer Factory
Creator

Stream Header

Used mainly for compatibility with existing serialization
protocols, to write out or consume a fixed header

Convenience static methods for fixed byte patterns

Usage example: StreamHeader

(example)

Class Table

Purpose
• To avoid sending lengthy class descriptor information by using short user-

defined sequences of bytes to represent certain classes

Uses
• Improve performance by potentially drastically decreasing message size
• Most effective when sending and receiving class hierarchies are known to

match
 Writing serialized data to a file, reading it back by the same program later
on
 Sending data between two identical nodes in a cluster

Usage example: ClassTable

(example)

Object Table

Purpose
• To allow certain object instances to be written as user-defined sequences of

bytes

Uses
• Remote references
• Space-efficient serialization of constants and singletons

Usage example: ObjectTable

(example)

Class Resolver

Purpose
• To allow hooks into the class descriptor reading and writing process

Uses
• To specify a ClassLoader policy to use upon unmarshalling
• To substitute class names on the fly
• To use shortened abbreviations for common class names

Equivalent to...

• ObjectInputStream.resolveClass() and resolveProxyClass()
• ObjectOutputStream.annotateClass() and annotateProxyClass()

• Other features not directly available in std. serialization

Usage example: ClassResolver

(example)

Object Resolver

Purpose
• Provide a means to substitute an object before marshalling or after it has

been unmarshalled.

Uses
• Sending a Serializable substitute for a non-Serializable object (substituting

on both ends)

• Interning Strings or similarly cached objects (on receiving end)

Equivalent to...

• ObjectInputStream.resolveObject()
• ObjectOutputStream.replaceObject()

Usage example: ObjectResolver

(example)

Externalizers

Purpose
• An Externalizer is an object which knows how to “externalize” a (possibly

non-Serializable) class

Uses
• Replace uses of java.io.Externalizable in order to:

 Prevent public exposure of the “readExternal” and “writeExternal”
methods

 Allow initialization of classes with final fields

• Provide the most efficient object serialization possible
• Integrate with other marshalling frameworks

 Example: Protocol Buffers

Externalizers (cont.)

Two ways to define an Externalizer
• By specifying a ClassExternalizerFactory in the configuration

 Returns an Externalizer given a Class argument

• Through use of the @Externalize annotation

 Specifies a class with a no-arg constructor, which implements
Externalizer

The Externalizer is sent with the class information
• Therefore, the Externalizer itself needs to be Serializable, Externalizable, or

have an Externalizer of its own

• Using an ObjectTable entry will also suffice

Usage example: Externalizers

(example)

Creator

Purpose
• To provide an object creation strategy

Uses
• Used to create uninitialized objects
• Two implementations are included

 ReflectiveCreator: uses reflection to attempt to locate and execute a no-
args constructor

 SunReflectiveCreator: create an object using the internal JDK API
reserved for serialization object creation (may not be available on all
platforms)

Usage example: Creator

(example)

Questions

Thanks!

All demos are available here:

http://anonsvn.jboss.org/repos/sandbox/david.lloyd/marshalling-demos
More questions? Visit the #jboss IRC channel on irc.freenode.net!