Collaborative best-effort real-time agents-a new paradigm

Abstract

This paper introduces a novel new class of network-based software agents, called CoBERT agents, because they are designed to cooperate in the best-effort maximization of global system objectives for a dynamic set of agents with evolving characteristics, while reliably accommodating real-time requirements, as well. Cooperation is achieved by supporting mobility within a subnet of processing nodes, morphability between sets of available workpoints, decision epochs for shifting to other nodes and/or workpoints, appropriate state-dependent functions for expressing decision options and implications, and an appropriate protocol for sharing this information with meta-control agents charged with continuously monitoring the network and making the best-effort decisions. The timely scheduling of hard and soft real-time activities is accomplished even though there is no a priori knowledge as to where the agent will run and with which additional CoBERT agents. Such capabilities are essential to future agent-based systems, as real-time grows pervasive and as dynamic sets of evolving agents populate our networks. A distributed integrated HIL simulator is used to introduce the real-time domain and to demonstrate how the CoBERT agent paradigm might be applied.