Agent Oriented Software Engineering with MESSAGE.
Jorge J. Gómez-Sanz, Juán Pavón · 2002
After reviewing existing AOSE methodologies, we have the impression that some more work is still needed, from a software engineering viewpoint, in order to apply them in an industrial context. For instance, the software development process is usually summarized in four or five steps for analysis and design, which is not realistic. Just taking a look at the Rational Unified Process [4] is enough to get an idea that the definition of an object-oriented methodology is not trivial. And we may agree that developing a Multi-Agent Systems (MAS) is more complex than a conventional object oriented application. It is true that in some cases, like MaSE [2], the presence of a tool, like AgentTool [7], simplifies the development process. However, this brings the risk to loss in flexibility. ZEUS [6] facilitates the development process fixing the target multi-agent system architecture. AgentTool sacrifices the flexibility in the design assuming a fixed set of diagrams mostly based on state-machine-like specifications similar to those of SDL [3]. When the MAS is a small one, when the problem is restricted to academic domain, such tools are really useful and a methodology of just a short number of steps sounds reasonable. However, an industrial development, involving a team of several engineers requires management of activities, a more detailed software development process, and the control of the cost involved in making a MAS work under real world workload. To deal with MAS development problems, specially in analysis and design phases, we need flexible tools and specification methods that permit researchers develop MAS as if they were developing conventional software. And this does not mean that there is a fixed target platform which has to be instantiated with some parameters. Target platform is important but should not suppose a constraint for a methodology. With this assumption, MESSAGE methodology [1] proposes to integrate well-proven object-oriented software engineering practices with agent technology results.