Visualization of the Carrot 2 System Thomas Laufert

Visualization of the Carrot 2 System Thomas Laufert Center for Architectures for Data-Driven Information Processing University of Maryland Baltimore County

Outline l l l l Introduction Background Related Work Carrot 2 Visualization • • Ontology Implementation Discussion Future Work Conclusion

Introduction l What is Carrot 2? • Distributed Multi-Agent Information Retrieval System l Why is Visualization desired? • Autonomy can lead to unpredictability and • confusion Is it necessary?

Introduction l Agent Visualization Problem l What’s Next? • Large amount of information • Limited real estate • Performance

Background l Debugging • Debugging multi-agent-systems can be difficult. l Visualization of communication patterns l Demonstrating • Is there method in its madness? • To show to large audiences the performance of the Carrot 2 system.

Related Work l Ndumu/ZEUS l Szekely – Punch Card System • Multiple domains • Separation of tasks to different tool sets • Punch card where each “hole” represents an • agent, different colors represent status of the agent. Scalability – tens of thousands

Related Work l JADE • Visualization incorporated into JADE • l environment Multiple tools; Control, debugging, and message visualization RETSINA MOCHA • Mobile agent-communication framework • Visualization incorporated into framework

Carrot 2 Visualization l Scaleable to tens of thousands of agents Separation of tasks l Event Ontology l • Queries • Meta-data

Event Ontology l Event Hierarchy

Event Ontology l Agent Registration l Query Routing • Agents must register with the ANS • When agents receive a query they route the query to determine what action to perform on it l Query Receipt • When an agent receives a query they throw an event.

Event Ontology l l Query Forwarding • The agent can forward the query to a more appropriate agent to handle the query Query Processing • Process the query against its collection • Query Processing has been performed and agent responds to the sender Query Response Receiving the Results of a Query • An agent receives the results of a recently processed query

Event Ontology l Other Events l How is this information useful in debugging? • Meta-data • Collection distribution

Implementation l l Java Application using Swing Logger Agent • Logs the events given to it from the agents in • • the system to a file Visualizer gets all information from the Logger agent Will not affect performance significantly

Implementation l l l Visualizer can work in an offline or online state JLists representing event history lists for each node in the system. Events differentiated by color of text in the JList

Implementation l l l l Agent Registration Query Routing Query Receipt Query Forwarding Query Processing Query Response Receiving Results

Discussion l How to determine what information to visualize • Limited Real Estate l Online and Offline visualization modes

Future Work l Augmented Event Ontology l Incorporate Logger agent as an component in the visualizer tool l Port Visualization tool to work in other systems

Conclusion l With large scale multi-agent-systems visualization is important to aid in: • Debugging • Understanding communication l Visualization makes it easy

References l l l JADE: http: //sharon. cselt. it/projects/jade RETSINA MOCHA: http: //www 2. cs. cmu. edu/afs/cs. cmu. edu/project/rets ina-22/website/index-04. html ZEUS: http: //www. btexact. com/projects/agents/z eus/