Agents for Datagrid

1
Agents for Datagrid

• Luciano Serafini Floriano Zini
• ITC-IRST
• Trento Italy
• Geneva, August 2000

2
Motivations

• Software development becoming more and more
  complex and expensive, because of increasing
  requirements in terms of
• temporal availability (24 hours x 7 days/week
  required even for the simplest Internet service),
• spatial availability (access from everywhere,
  portable functional sophistication, ...),
• integration with external systems (legacy as well
  as new),
• support for heterogeneous platforms (hardware and
  software).
• Thus, there is a need for novel approaches to
  software engineering, including multi-agent
  systems.

3
Multi-Agent Systems Overview

• Multi-agent systems (MAS) are collections of
  intelligent processes (agents) cooperating
  towards some goals. General characteristics
  include
• distribution over intranets or the Internet
• autonomy agents take simple decisions, run
  largely unsupervised
• reactivity agents respond to changes in the
  environment
• pro-activity agents have goals of their own
  changing with environment,
  seem to take the initiative''
• sociality agents easily interact with other
  systems and humans MAS organized according to
  some social paradigm, e.g. teams achieving joint
  goals.

4
Evaluation ofExisting Agent Technologies
5
JACK Intelligent Agents

• A product by Agent Oriented Software Pty.
  Ltd.(http//www.agent-software.com)
• 2.0 released November 1999
• Based on companys RD
• third generation agent system
• Component-based approach
• core architecture and capability for developing
  running distributed software agents
• allows for variety of types of software agent to
  be layered on top of base kernel, from simple
  agents to teams of intelligent agents

6
JACK features

• Design philosophy for integration, re-use and
  re-engineering
• Low resource requirements, designed to handle
  hundreds of agents running on low-end hardware
• Additional functionality is provided in the form
  of plug-in components, depending upon
  requirements
• Language specification and modular design allow
  easy extension for new agent models

7
JACK features

• Belief-Desire-Intention Agent Model
• advanced programming capability
• with type-safeness
• multi-threading with safety net
• Capability concept
• allows the agent elements to be structured
• provides encapsulation of functionality
• declares interactions between encapsulated
  functionalities
• introduces sound software engineering practice to
  agent programming

8
Benefits of JACK

• Rapid development of distributed systems composed
  of agents
• Full access to Java
• Light-weight implementation
• High-level procedural logic within a fully
  fledged object-oriented environment
• Applications include
• business procedures
• workflows
• simulation of human reasoning

9
Entirely written in Java

• Portability
• capable of running on any system on which Java is
  available
• Access to all Java capabilities, including
• multiple threads possibly running on multiple
  CPUs
• platform independent GUIs
• third party libraries such as JDBC
• Easy integration with external packages using
  standard infrastructure, e.g., CORBA, RMI

10
The BDI Agent Model
Human
Belief, Desire, Intentions Agent
Beliefs - database of perceived world knowledge
Beliefs - perceived understanding of the world
Goals or desires
Goals or desires
Execution Engine
Intentions - currently executing plans
Pre-compiled plans
Accumulated experience and behaviours
11
JACK BDI Execution
observe change
intentions
beliefs
desires
observe "environment"
step intention
messages
12
JACK BDI Execution
intentions
beliefs
desires
13
JACK BDI Execution
observe change
intentions
beliefs
desires
observe "environment"
messages
14
JACK BDI Execution
observe change
intentions
beliefs
desires
observe "environment"
step intention
messages
15
JACK Programming Concepts
event
agent
plan
database
capability
has
uses
16
Agents

• A type
• agent AgentType extends Agent ...
• Encapsulates knowledge and behaviour (databases,
  events and plans)
• Reacts to events -- performs tasks
• Receives messages -- performs services
• Interfaces other system components
• GUI, back-end, other processes

17
Capabilities

• A type
• capability CapabilityType extends Capability
  ...
• Allows agent elements to be structured
• Provides encapsulation of functionality
• Enforces declared interactions between
  encapsulated functionalities
• Introduces sound software engineering practice to
  agent programming

18
Events

• A type
• event EventType extends Event ...
• Provides the type safe connections between agents
  and plans
• both agents and plans must declare the events
  they handle as well as the events they post or
  send
• Base type defines plan processing behaviour
• Event, MessageEvent, BDIGoalEvent, ..

19
Databases

• A type
• database DatabaseType extends ClosedWorld ..."
• Defines knowledge capability (relational
  modelling)
• closed world / open world semantics
• Each database definition results in two Java
  classes
• one (DatabaseType) for the relation, and
• one (DatabaseType__Tuple) for the tuples of the
  relation

20
Plans

• A type
• plan PlanType extends Plan ..."
• Defines context dependent responses to event
  occurrences
• Plan processing succeeds or fails
• Reasoning methods
• requires each step to succeed
• each step processed atomically
• Meta-level plans to handle plan choice

21
Agents for Data Managements

• Overhearing Agents They overhear on a dialog
  between a community of agents, and they provide
  such information to the suggesting agents.
• Suggesting Agents They suggest alternative
  and/or additional actions and information to an
  agent by accessing to its dialogues information.

22
Architecture Example

• Agents dialogue

Client
Server
Overhearing Agent
Query suggesting agent
Active Caching agent
Mirror suggesting agent
23
Overhearing Functionalities

• Listening dialogues
• Storing dialogues
• Analyzing dialogues
• Reporting on dialogues
• . . .
• NB Overhearing agents can be positioned in any
  communication point of the DataGrid

24
Overhearing Technologies

• Standard technologies (DB, . . . ) for storing,
  and integrating with other DataGrid components
  (Monitoring, )
• Formal reasoning for discovering query
  equivalence, query containment, etc.
• Case base reasoning for dialogue analysis, query
  similarities, etc.

25
Query Suggesting Functionalities

• Query optimization
• Q(X1, . . . , Xn) ? Q(Y1, . . . , Ym)
• Query estimation
• Q(X1, . . . , Xn) ? Answer time
• Answer size
• . . .

26
Query Suggesting Technologies

• Formal reasoning on queries for
• Query rewriting
• Query containment
• Query classification
• Case Based Reasoning on queries for
• Query estimation (time, size, )

27
Active Caching Functionalities

• Caching Maintain a local copy of the results of
  the queries submitted by the users.
• Pre fetching Predict the future queries of the
  users and pre cache the results of these queries
• . . .

28
Active Caching Technology

• Standard caching technologies For storing the
  results of the queries performed by the user.
• LearningFor building a user model, which is
  necessary to predict its queries.
• Formal reasoning To state when the result of a
  certain query is already present in the cache, as
  an answer of a previous query.

29
Mirror Suggesting Functionalities

• Mirror suggestion Suggest to the user
  alternative and more convenient client where to
  submit his query.

30
Mirror Suggesting Technologies

• Standard technologies, for describing the
  distribution and the duplication of the data in
  (a subset of) the whole DataGrid
• Metadata For describing the content of the nodes
  of the DataGrid
• . . .

31
Final Remarks

• Suggesting agents can be defined for the server
  side, and for all the other components of the
  system. Many other kind of suggesting agents can
  be defined.
• The ones provided in this presentation are
  examples.
• A detailed analysis of the domain and of the user
  requirements is necessary in order to precisely
  define the Suggesting Agents.
• This architecture allows for an incremental
  implementation and testing of suggesting agents.