ABLE: A toolkit for building multiagent autonomic systems

1
ABLE A toolkit for building multiagent autonomic
systems

• Bigus, Schlosnagle, Pilgrim, Mills, Diao

2
Review of agent --from wikipedia

• a complex software entity that is capable of
  acting with a certain degree of autonomy in order
  to accomplish tasks on behalf of its user. An
  agent is defined in terms of its behavior. (vs
  method, attributes for methods)
• distinguish agents from arbitrary programs
  reaction to the environment, autonomy,
  goal-orientation and persistence.
• Distinguishing agents from objects
• flexible behaviour reactive, proactive, social
• at least one thread of control but may have more
  

3
What is ABLE

• Purpose of ABLE toolkit
• fast, reusable, and scalable architecture for the
  construction of intelligent software components
  and agents

4
What is ABLE

• Agent building and learning environment
• Component base approach Java base
  toolkit-portability, oo, native to multi-thread
  et al., hybrid intelligent agent application
• Java beans AbleBeans component library of data
  access, machine learning, machine reasoning, and
  optimization algorithms
• Sun Microsystems defines them as "reusable
  software components that can be manipulated
  visually in a builder tool".
• Platform for distributed computing

5
AbleBean connections

• Java beans deployed as Java components/agents
• Three methods for AbleBean connection
• Dataflow
• In/out buffer
• DAG

6
AbleBean connections

• Event Connections
• AbleEvent synchronous and asynchronous event
• Each AbleBean has event queue on which receives
  AbleEvent notification.
• Data event (evenID), data object, action event
  (action string)
• Flexbility vs hard coded method calls
• Property Connections
• Synchronize two properties in two beans
• Changes via setter method
• AbleBean use serialization for persistence

7
ABLE agent framework
8
Able class

• Event Class
• Dialog between two agents can be supported by
  exchanges of AbleEvents
• An agent broadcast its response to an event to
  multiple agents
• Object Class
• Base implementation of AbleBean interface
• Agent Class
• AbleDefaultAgent

9
The ABLE agent
10
ABLE component library of beans

• Databean
• Data access, transformation
• Learningbean
• Learning algorithms
• Rulebean
• AbleRuleLanguage, rule-base knowledge

11
Function Specific Agents

• Genetic search agent
• Neural classifier agent
• Neural clustering agent
• Neural prediction agent
• Script agent
• Java script agent

12
ABLE Development Tools

• ABLE Agent Editors
• Different views
• Graphical construct agents
• How it works? Load beans from jar files
• ABLE inspector windows
• display bean data members and state

13
ABLE agent platform

• Distributed agent platform
• RMI
• Communication between agents
• Conform to FIPA 97 (Foundation for Intelligent
  Physical Agents)
• JAS (Java Agent Services)

14
(No Transcript)
15
Case1 System Admin

• IBM eServeriSeries
• Monitor
• Cpu utilization, workload
• Disk utilization

16
(No Transcript)
17
Case2 Diagnostic application

• Sever diagnostic data collection, problem
  determination and problem source identification
• Autotune Agent contains one or more
  AutotuneController beans that provide control
  strategies and one or adaptor beans (interface to
  the target)

18
Case2 Diagnostic application

• ARL time series analysis, reasoning and data
  analysis
• Diagnostic manager coordinate the activities of
  individual diagnostic agents

19
Case3 Autotune agent for Apache

• Turning of
• MaxClients
• Adjust according to the dynamics of the server
• KeepAlive
• Larger keepAlive value is used to decrease the
  CPU level

20
Subsumption agent and Autonomic agent

• AbleSubsumptionAgent
• reflexive, reactive and adaptive
• AbleAutonomicAgent
• Cooperate and compete to take control of
  intelligent system make response
• Behaviors Basic reflexive, complex instinctive
  and learn from interaction with the world

21
(No Transcript)
22
Comments

• ARL-mix rule-based reasoning with neural network
  and other machine learning algorithm.
• Neural trained agents

23
Discussion

• About the distributed agent platform, it can be
  used to achieve emergence (?)
• A practical autonomic system would be composed of
  many controller agents, distributed across many
  computer systems
• Use of an agent structure with well-defined
  functional components, where those components
  themselves are multiagent systems