AspectJ Based Implementation of Dynamically Pluggable Filter Objects in Distributed Environment

1
AspectJ Based Implementation of Dynamically
Pluggable Filter Objects in Distributed
Environment

• Rushikesh K. Joshi, Neeraj Agrawal
• Department of Computer Science and Engineering
• Indian Institute of Technology Bombay, India
• Contact Rkj_at_cse.iitb.ac.in

2
The Filter Object Model based on Interclass
Filter Relationship

3
Properties of Filter Objects

• Filter Object Specification decoupled from
  Client/Server class specifications
• Filter Objects are instance of classes
• Transparency client and server code unaware of
  filter objects
• Filtering of upward and downward messages
• Dynamic Pluggability specify, create, plug,
  reconfigure at runtime
• Chaining and Grouping

4
Key Features of the AspectJ Implementation of
Filter Objects

• An Aspect specification - To model filter objects
• Calls delegated to a filter pipe pointcuts and
  around advice
• Filter Pipe - Organizes the chain of filter
  objects, which may themselves be distributed
• Filter pipe is a remotely available object
• Maintains Mappings from client-server pairs to
  vectors of filter objects (upfilters and
  downfilters)

5
The Solution

Aspect
FilterObject upFilter1 () downFilter1 () ..
FilterObjectN upFilter1 () downFilter1()
FilterPipe upFilter1 () downFilter1 (..)
FilterObject1 upFilter1 () downFilter1()
FOs
..
For all f in FOs f-gtupFilter()
6
Development Process

• Base Code ? Filter Class Abstraction ? Base Code
  Compilation with Abstraction (needed if
  environment is not filter aware) ? Base Code in
  execution ? Filter Object Development may be
  developed as late as this point after their
  abstraction is known ? Inject filters into the
  system, reconfigure as programmed for

7
Filter Binding Types
One-to-One
One-to-Many
Many-to-Many
Many-to-One
8
Typical Deployments

• One-to-One Client Side/Server Side
• Low network overhead
• One-to-Many Client Side
• Client side deployment more efficient since all
  filtered messages originate from one client
• May-to-One Server Side
• Server side deployment more efficient since all
  filter messages are against one server
• Many-to-Many Possibly on an Independent
• Machine
• Depends on network overheads, load
• Per Class/Per Object/Per Machine Filter Pipes

9
Conclusions

• Proposed a scheme for implementation of
  dynamically pluggable and distributed Filter
  Objects on top of AspectJ
• Requires base code to be filter aware when in
  execution
• Supports rich possibilities of configuring filter
  objects
• Brought out the need for a careful selection of a
  Deployment Strategy

10
Issues put forward for Discussion

• Is the requirement of Filter Aware Base Code a
  desirable property?
• Consider Static vs. dynamic awareness
• Is there an advantage in turning Programming
  Environments themselves Filter Aware
• What is the right granularity of filter
  specifications and is there synthesis? per
  process, per object, per message, per machine,
  per class, per client ..chains,groups..
• Is there separation of high level abstractions of
  filters from their low level generalizations? How
  efficient can the mappings from former to latter
  be?