This phase influences and is influenced by the Agent Behaviour Description phase as a double level of iteration occurs between them. The Agent Structure Definition phase produces several class diagrams logically subdivided into two views: the multi-agent and the single-agent views. In the former, we call attention to the general architecture of the system and so we can find agents and their tasks in it. In the latter, we focus on each agent's internal structure, revealing all the attributes and methods of the agent class along with its inner classes (i.e., its tasks)..
At this stage, one class diagram represents the MAS as a whole (see Figure 1). The diagram shows classes, each symbolizing one of the agents identified in the A.Id. phase. Attributes compartments can be used to represent the knowledge of the agent as already discussed in the Communication Ontology diagram, whereas operations compartments are used to signify the agent's tasks.
Figure 1: The Multi-Agent Structure Definition diagram for the bookstore case study
Here one class diagram (see Figure 2) is used for each agent to illustrate the agent's interior structure through all of the classes making up the agent, which are: the agent's main class together with the inner classes identifying its tasks. At this point, we set up attributes and methods of both the agent class (e.g. the constructor and the shutdown method required by FIPA-OS environment) and the tasks' classes (e.g. methods required to deal with communication events such as the handleRequest method of the IdleTask invoked when the agent gets a request communicative act).
The result of this stage is to obtain a detailed structure of the software, ready to be implemented almost automatically.
Figure 2: The Single-Agent Structure Definition diagram for the PurchaseManager agent
Next phase: Agent Behavior Description
Previous phase: Protocol Description
Home: PASSI homepage