On the Encapsulation and Reuse of Decentralized Coordination Mechanisms: A Layered Architecture and Design Implications

Overview & Synopsis

The effective and reliable coordination of agent activities is a momentous problem for Multi–Agent System (MAS) developers. Particularly challenging is the decentralized coordination of agents that enables systems to exhibit self–organization. Natural phenomena typically serve as design metaphors and developers apply Decentralized Coordination Mechanisms (DCMs) that have been inferred from biological, physical or social systems. This paper addresses the utilization of DCMs as reusable software components.Current development practices give little guidance for DCM selection and force developers to manually design, implement and tune mechanism parameters ad hoc, leading to highly specialized algorithms. Here, we propose a layered software architecture that encapsulates DCMs in (multiple) coordination spaces. A generic, annotation–based interface allows to separate agent coordination from agent functionality, therefore enabling DCM reuse and facilitating application (re)designs, i.e. mechanisms exchange and parameter adjustments. Implications for development procedures are discussed and the application of the layered architecture is exemplified in a resource allocation case study.