Abstract

"Component-based approaches to elaborate software must deal with the fact that in practical settings, components information may be incomplete, imprecise and uncertain, and requirements may be likewise. Architects wanting to evaluate candidate architectures regarding requirements satisfaction need to use whatever information be available about components, however imperfect. Imperfect information can be dealt with using specialized analytical formalisms, such as fuzzy values for imprecision and rough sets for incompleteness; but if used, evaluations need to compare and rank using non-scalar, non-symbolic values. This article presents an approach to systematically describe components' imperfect information, and to evaluate and rank whole component assemblies, by using credibility values-based ""support scores"" that aggregate imperfect information about requirements, mechanisms and components. The approach builds on the Azimut framework, which offers progressive refinement of architectural entities via architectural policies, architectural mechanisms, components, and component assemblies. An example of the proposed approach and ""what-if"" analysis are illustrated. © 2006 Springer-Verlag."