Resilience through Dynamic Reconfigurations in Agent Systems Ilya

  • Slides: 15
Download presentation
Resilience through Dynamic Reconfigurations in Agent Systems Ilya Lopatkin Newcastle University, School of Computing

Resilience through Dynamic Reconfigurations in Agent Systems Ilya Lopatkin Newcastle University, School of Computing Science

Multi-Agent System (MAS) Resource Active agents Resources Agent Communication Agent Resource 2

Multi-Agent System (MAS) Resource Active agents Resources Agent Communication Agent Resource 2

Multi-Agent System (MAS) Agent Active agents Passive agents Agent Communication Agent 3

Multi-Agent System (MAS) Agent Active agents Passive agents Agent Communication Agent 3

Threats and reconfiguration in MAS § § § Agent is unavailable Abnormal behaviour Disconnection

Threats and reconfiguration in MAS § § § Agent is unavailable Abnormal behaviour Disconnection Inadequate quality of data Inadequate Qo. S Lack of required characteristics Agent ? ? Agent Questions concerned: § how to find appropriate components to use after failures? § which of them to choose? 4

The place of the search mechanism in FT Error detection Damage confinement and assessment

The place of the search mechanism in FT Error detection Damage confinement and assessment Fault treatment and continued service Error recovery Search mechanism * T. Anderson, P. A. Lee. Fault Tolerance: Principles and Practice. Prentice Hall, London, 1981 5

Scope of work Error detection Search Location addresses ØRequest Reconfiguration List of locations Ø

Scope of work Error detection Search Location addresses ØRequest Reconfiguration List of locations Ø Searcher Criteria 6

CAMA abstractions Agent Middleware Location Agent Middleware Platform Device 7

CAMA abstractions Agent Middleware Location Agent Middleware Platform Device 7

Evaluation of locations Agent evaluates Location Searcher produces Value* * G. Di Marzo Serugendo,

Evaluation of locations Agent evaluates Location Searcher produces Value* * G. Di Marzo Serugendo, J. Fitzgerald, A. Romanovsky, and N. Guelfi. A Metadata-Based Architectural Model for Dynamically Resilient Systems. In 22 nd ACM Symposium on Applied Computing, Seoul, Korea, March 1115, 2007. pp 566 -573. ACM, 2007. 8

Sharing values among agents uses Criterion uses Agent B uses Value A 2 Agent

Sharing values among agents uses Criterion uses Agent B uses Value A 2 Agent A reads from Location 1 saves to evaluates Location 2 Value A 2 9

Aggregating values Value PQ 2 Location 1 aggregates into contains Value Q 2 reads

Aggregating values Value PQ 2 Location 1 aggregates into contains Value Q 2 reads Q 2 saves PQ 2 Agent A Location 2 reads P 2 Location 3 contains Value P 2 10

Criteria Application-specific Evaluate, aggregate, and compare values May include any resilience criterion Examples: Availability.

Criteria Application-specific Evaluate, aggregate, and compare values May include any resilience criterion Examples: Availability. Value: estimated time per week/month/year Connection properties, latency. Value: average time in ms Number of failures. Value: integer Quality of service. Value: some complex structure 11

Threats and reconfiguration in MAS meta-data § § § Agent is unavailable Abnormal behaviour

Threats and reconfiguration in MAS meta-data § § § Agent is unavailable Abnormal behaviour Disconnection Inadequate quality of data Inadequate quality of service Lack of required characteristics Agent ? ? Agent Questions concerned: § how to find appropriate components to use after failures? § which of them to choose? Agent meta-data 12

Advantages Works for any type of components Leads to <reconfiguration type you've just remembered>

Advantages Works for any type of components Leads to <reconfiguration type you've just remembered> Any resilience criteria Autonomous agents High scalability 13

Problems First wave of agents Too many locations Lack of values Different treatment of

Problems First wave of agents Too many locations Lack of values Different treatment of the same values Need for a general ontology to represent values Hidden stigmergy may lead to unpredictable behaviour 14

Thank you!

Thank you!