This paper presents an innovative multi-agent system development approach called mission-based on-demand agent generation. This approach allows agents to be dynamically composed at run-time and most importantly, only when needed. Such an approach is different from the conventional one, where agents are generally composed at design time. Our model of a mission allows the MAS to be suspended and resumed at later stage at the same or different location. We present the formal model of the mission, the strategy to execute the mission and the architecture of the prototype system called eHermes. Finally, we report the experimental results that shows that eHermes handles the load satisfactorily and performs the run-time optimization well.
Glenn T. Jayaputera, Arkady B. Zaslavsky, Seng Wai