Towards Robustness Analysis of Component-based Systems built on Imperfect Information

Component selection is a key step to build component-based software systems, but in practice available components’ information is usually imprecise, incomplete, uncertain and/or unreliable; good components may pass unnoticed. This article presents the Azimut approach to generate alternate assemblies of Off-the-Shelf Component as candidate solutions at early project stages. Azimut deals explicitly with imperfect information, by modeling information on components as (score, credibility) pairs where credibility is assessed by evaluators themselves, and fitness of assemblies as solution candidates is evaluated with aggregated scores. This separation of evaluation and credibility enables robustness analysis of a solution set regarding the quality of available information, i.e. determining how sensitive is received architectural advise to changes in information. Thus, architects can explicitly tune their willingness to use imperfect information, possibly deciding to use only assemblies with high robustness, or allocating extra effort to improve some key components’ information. A detailed example illustrates the approach and its use to support early architectural decisions.