Abstract
Architecture-based software reliability analysis can provide early identification of critical components which can then be targeted for cost-effective reliability improvement of the application. However, an important challenge in conducting this analysis early in the life cycle is that it is nearly impossible to estimate the architectural and component parameters with certainty. The issue of estimating software application reliability in the presence of uncertain component reliabilities has been addressed in the previous research. In this paper we consider the estimation of software reliability in the presence of architectural uncertainties. We present a methodology to estimate the confidence levels in the architectural parameters using limited testing or simulation data based on the theory of confidence intervals of the multinomial distribution. The sensitivity of the system reliability to uncertain architectural parameters can then be quantified by varying these parameters within their confidence intervals. The illustration of the methodology using a case study indicates that the impact of the uncertainty in a given architectural parameter on the overall application reliability is determined by the inherent branching behavior of the application and the component reliabilities.