1 / 17

Coordination Artifacts in Multi-Agent Systems

Coordination Artifacts in Multi-Agent Systems. April 19, 2005 IEEE KIMAS 2005 Sarah Siracuse, John Zinky, Richard Shapiro Ssiracus@bbn.com, jzinky@bbn.com, rshapiro@bbn.com. Agenda. Motivation for Coordination Artifacts in MAS Coordination Artifacts: Designs & Benefits

hollye
Télécharger la présentation

Coordination Artifacts in Multi-Agent Systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Coordination Artifacts in Multi-Agent Systems April 19, 2005 IEEE KIMAS 2005 Sarah Siracuse, John Zinky, Richard Shapiro Ssiracus@bbn.com, jzinky@bbn.com, rshapiro@bbn.com

  2. Agenda • Motivation for Coordination Artifacts in MAS • Coordination Artifacts: Designs & Benefits • Separation of Function: Coordination logic vs. Domain logic • Implementation of Coordination Artifacts using Cougaar • Works well in Tightly-coupled Systems • Performance Analysis: QoS Opportunity • Conclusions

  3. Coordination observations MAS application Cougaar agent architecture ~1000+ agents on ~100 hosts Many different kinds of implicit coordination in heterogeneous systems Coordination implementation Mixed in with domain logic Spans lots of places in the code Coordination Artifact Separates coordination implementation from domain logic Distinguishes between various kinds of coordinations Has state Motivation for Coordination Artifactsin MAS Typical Agent Control Society Manager Aggregate Summarize Disseminate Synchronize Controller Peer Translate Collect Sensor

  4. Objective Coordination (Outside Agent) Coordination encapsulated Outside domain logic Environment-based Mediated communication e.g. Ant trails Subjective Coordination (Inside Agent) Coordination mixed in with domain logic Dialog-based Direct Messaging e.g. TCP/IP, Instant Messaging,FIPA Agent Communication Language CAs Separate Coordination Activityfrom Domain Processing Agent Agent Agent Agent Agent Agent Agent Agent Agent Agent CAs

  5. CAs are First Class Entitiesthat coordinate Interaction between Agents Defines roles • Coordination Artifacts: CAs • Are first-class entities in MAS • Define explicit roles for role-players • Offer shared state between the role-player & the CA • Coordinate behavior among role-players • Have distributed implementation Coordination Artifact (CA) Agent Agent Role-players Shared state Agent Agent

  6. CAs Unify Agent-to-Agent andAgent-to-Environment Communication Persisted Storage Other Agents Physical Environment Agent Sensor Coordination Inter-Agent Coordination Client CA Server CA Non-Agent Systems

  7. Agenda • Motivation for Coordination Artifacts in MAS • Coordination Artifacts: Designs & Benefits • Separation of Function: Coordination logic vs. Domain logic • Implementation of Coordination Artifacts using Cougaar • Works well in Tightly-coupled Systems • Performance Analysis: QoS Opportunity • Conclusions

  8. Cougaar Components ImplementAd-hoc Coordination Remote Agents Agent Net Physical Environment Agent Domain Logic Inter-agent Messaging Components Sensor Comp Agent Blackboard Sensor Plugin Comm Plugin Client Plugin Server Plugin Client Libraries Server Libraries Non-Agent Systems

  9. Distributed Coordination ArtifactsLayered Over Cougaar Components Receptacle Facet Receptacle Facet CA Agent A Agent B Rule Engine Rule Engine Fact Base Fact Base Fact Fact Black- board Black- board Relay Relay Logic Provider Message Transport Message Transport Logic Provider RMI Host A Host B

  10. Tightly coupled (Ideal CA applications): Long Term Relationships Group relationships Push meta-data in anticipation of need E.g. Cougaar with Coordination Artifacts Coordination Artifacts work bestin Tightly-Coupled Systems Typical Agent Control Society Manager Aggregate Summarize Disseminate Synchronize Controller Peer Translate Collect Sensor • Loosely coupled (Bad fit for CAs): • Transient Relationships • Pair relationships • Pull meta-data when needed • E.g. Web-Services

  11. Coordination Performance Depends onUnderlying Topology and Resources Coordination Task M Tick Sync Coordination S S S S …

  12. Coordination Performance Depends onUnderlying Topology and Resources Coordination Task M Dual Procesor Dual Processors Tick Sync Coordination S S S S … WAN M Resources & Roles Single processor 2.8GHz Single Processor Dual processors 2@2.0GHz Distant slave (31 hosts) Distant master (31 hosts) WAN S

  13. Coordination Performance Depends onUnderlying Topology and Resources Topology M Flat Coordination Task S S S S S S M Dual Procesor Dual Processors Tick Sync Coordination Tree M T T Chain S S S S S S S S … WAN M Flat Tree Chain M T T T T S T Resources & Roles Single processor 2.8GHz Single Processor Dual processors 2@2.0GHz Distant slave (31 hosts) Distant master (31 hosts) WAN S

  14. Coordination Performance Depends onUnderlying Topology and Resources Topology M Flat Coordination Task S S S S S S M Dual Procesor Dual Processors Tree M T T Chain S S S S S S S … WAN M Flat Tree Chain M T T T T S T Resources & Roles Single processor 2.8GHz 27 25 17 Single Processor Dual processors 2@2.0GHz 35 22 13 Distant slave (31 hosts) 6.1 2.5 0.8 Distant master (31 hosts) 0.6 2.5 0.8 WAN S Performance (ticks/second) Tick Sync Coordination S

  15. Coordination Performance Depends onUnderlying Topology and Resources Topology M Flat Coordination Task S S S S S S M Dual Procesor Dual Processors Tick Sync Coordination Tree M T T Chain S S S S S S S S … WAN M Flat Tree Chain M T T T T S T Resources & Roles Single processor 2.8GHz 27 25 17 Single Processor Dual processors 2@2.0GHz 35 22 13 Distant slave (31 hosts) 6.1 2.5 0.8 Distant master (31 hosts) 0.6 2.5 0.8 WAN S Performance (ticks/second)

  16. QoS Adaptation via CAs Topology M Flat S S S S S S Dual Procesor Dual Processors Tree M T T Chain S S S S WAN M Flat Tree Chain M T T T T S T Resources & Roles Single processor 2.8GHz 27 25 17 Single Processor Dual processors 2@2.0GHz 35 22 13 Distant slave (31 hosts) 6.1 2.5 0.8 Distant master (31 hosts) 0.6 2.5 0.8 WAN S Performance (ticks/second) CAs can dynamically change the topology as the network and/or the roles change.

  17. Conclusions • CAs present a formal model for coordinated communication • Blackboard-based, not FIPA message-based • Ease use of blackboard-based MAS • Unify Blackboard interfaces, including Web Services • Correlate multiple changes to blackboard objects • Partition the blackboard for domain and system reasons • Separation of Coordination and Domain processing • Make the intermediary a first-class entity • Place to add QoS-adaptation • Future Work • Might facilitate reuse or composability of coordinations • Might examine them in off-line analysis • Might support code generation

More Related