Detection of undesirable communication patterns in multi-agent systems

Two tool prototypes that measure the proposed metrics:Performs the measurement after the execution of the MAS from its logsPerforms the measurement during the execution using Application Programming Interfaces (API).

Development of MAS communication metrics, testing

Metrics related to the Quality of Service (QoS) – with its corresponding classification rules for each:Agent’s performancePerformance of a MASRole performance

Empirical case studies involving human actors interacting with the MAS: A customer agent needs to locate a suitable agent to perform a service on its behalf, and once located, it has to ask for it. Crisis management case study described by: García-Magariño et al. (2009)

Through case study

Aim: efficiently find and describe communication patterns that should be avoided in multi-agent systems (MAS) and identify the agents involved in these.Main goal is the measurement and improvement of communication architectures, instead of the balance of the workload of the components.

There is a lack of a large set of case studies to determine which metrics and contexts are useful when testing the communication designs of MASThe metrics have only been applied to post-mortem analysis, even if monitoring could be made during runtime.

Metrics and classification rules allow to identify several typologies of undesirable communication categories in MAS: overloader, overloaded and isolated agents.Engineers can improve the performance of the MAS by improving the communication design, being guided by the presented metrics.

Two MAS case studies considered:Crisis-management: poisonous material has been released in a city affecting part of its population. State of emergency declared.small MAS supports emergency servicesCases related to: coordinating citizens, warning citizens of affected locations, keeping emergency services informed( “test-bed” for developing the approach and definition of its componentsMAS to buy cinema ticketsComplex MAS (several classes of agents and roles)Cases related to: making plans, buying tickets, and buying extras.( validates the scalability of the approach (complex MAS).