Reliability Analysis of Software Architecture Evolution

Abstract

Software engineers and practitioners regard software architecture as an important artifact, providing the means to model the structure and behavior of systems and to support early decisions on dependability and other quality attributes. Since systems are most often subject to evolution, the software architecture can be used as an early indicator on the impact of the planned evolution on quality attributes. We propose an automated approach to evaluate the impact on reliability of architecture evolution. Our approach provides relevant information for architects to predict the impact of component reliabilities, usage profile and system structure on the overall reliability. We translate a system's architectural description written in an Architecture Description Language (ADL) to a stochastic model suitable for performing a thorough analysis on the possible architectural modifications. We applied our method to a case study widely used in research in which we identified the reliability bottlenecks and performed structural modifications to obtain an improved architecture regarding its reliability.