From m-GAIA to Grasshopper: Engineering Mobile Agent Applications

Abstract There is a need for agent oriented software engineering (AOSE) methodologies that support the conceptual modelling of mobile agent systems. In this paper, we present m-GAIA, our extension to the GAIA methodology for modelling mobile agent systems. m-GAIA incorporates explicit constructs to perform the analysis and design of multiagent systems which include mobile agents. We also present our experiences in mapping the conceptual models developed in m-GAIA to an implementation using the Grasshopper mobile agent toolkit. 1. Introduction A property of agents is the ability to move from one host to another to perform computations. While not all agents require such a property, agent mobility has been recognized as beneficial in a number of applications, including information-centric applications [3] and a sizeable number of mobile agent toolkits (http://mole.informatik.uni-stuttgart.de/mal/preview/preview.html) have been developed for mobile multiagent applications. While agent mobility may not be needed in all multiagent systems, we contend that it can be disadvantageous if it is a concept only confined to (and considered in) the implementation phase, or as an after-thought in engineering multiagent systems (e.g., used for optimising applications). The advantages of agent mobility might remain unexploited, adding mobility to agents after the system implementation can be problematic, or development of multiagent systems where some agents can be mobile might proceed in an ad hoc manner, if agent mobility is not considered in the earlier phases of agent-oriented software development. Whether agents are mobile or stationary does have an impact on the architecture and required components of a multiagent system, and how they can be made so, can be an important factor in deciding which agent toolkit to use for implementation. The why and how of agent mobility, or its dismissal, should be considered, and if needed, its rationale documented at design time. Such consideration of agent mobility should also support toolkit independence (as far as the required functionality will allow). On the other hand, there will some applications where agent mobility is obviously useful and more efficient, but an implementation in some mobile agent toolkit without the auspices of an integrated agent-oriented analysis and design is clearly less than ideal. What is the current state of the art in methodologies for the engineering of multiagent systems where some agents are mobile? Several approaches (or methodologies) for addressing the analysis and design phases of agent-oriented software have been reviewed [7]. Our analysis of these