Architecture-based Reliability Analysis With Uncertain Parameters

Architecture-based reliability analysis has gained prominence in the recent years as a way to predict the reliability of a software application during the design phase, before an investment is made in any implementation. To apply this analysis, the parameters comprising the architectural model must be estimated using the limited data and knowledge available during the design phase. These estimates, as a result, are inherently uncertain. Contemporary approaches, however, do not consider these uncertainties, and hence, may produce inaccurate reliability results. This paper presents a Bayesian approach to systematically consider parametric uncertainties in architecture-based analysis. The novelty of this approach lies in determining credible intervals for the model parameters as a function of their posterior distributions. By leveraging these intervals, we illustrate how to: (i) quantify the impact of uncertainty in a specific parameter on the system reliability estimate; (ii) evaluate when a sufficient amount of data has been collected to reduce the uncertainty to an acceptable level; and (iii) assess the impact of prior knowledge regarding the parameters in improving the system reliability estimate.