View composition in multiagent architectures

Our experience in research and development on a multi-agent architecture for an industrial Automatic Guided Vehicle Transportation System (AGVTS) reveals a crucial problem w.r.t. the specification of software architectures. In particular, there is a lack of support for relating and composing multiple views in architectural descriptions. This lack hampers changeability, as it prevents separating a concern while maintaining a good overview of the architecture as a whole. In this paper, we argue for explicit support in relating and composing architectural views. We propose three types of relations that allow specifying and composing structural views. To validate and evaluate our work, we extend a general-purpose architectural description language, xADL. We assess the usability and consequences of structural view relations in the context of the multi-agent architecture for the AGVTS. We learn that view relations and composition improve changeability of a multi-agent architecture by offering explicit support for (1) separating concerns in separate views and (2) composing the overall architecture from these separate views. The tradeoff for this improved changeability is the additional cost of documenting relations between architectural views. Nevertheless, documenting view relations give rise to architectural descriptions that are more expressive and less ambiguous.