| A Software failure, especially for large-scale software, can lead to a big catastrophe in many application domains. Therefore, reducing the chances of software failure and assuring software quality is highly desirable. Software reliability, a measure that estimates the probability that software will not cause the failure of a system for a specified time under specified conditions, thus is one of the key quality attributes. With the increasing sophistication of software infrastructures, the objective of this thesis is to model the reliability of large-scale software taking into account the architectural characteristics and complexity.; Traditional reliability models and approaches are subject to address either no system structures using a black-box approach or only simple homogeneous architectures using a white-box approach. Therefore, we proposed an architecture-based model, in which system structures are classified into architectural styles, in order to model the heterogeneity of different configurations and constraints. Our model extends the traditional Markov-based reliability models, a white-box approach, to ease the modeling of the interrelationships among components, and to assist in prioritizing the component improvement sequence.; However, the extension of traditional Markov-based models limits the modeling to only history-independent stochastic software behaviors: a problem resulting from the assumption that a software process follows a Markov process. Our model removes this fundamental barrier to address execution history, and both probabilistic and deterministic software behaviors, utilizing grammar rules and the structures of binomial trees without the necessity of history keeping. The grammar rules tackle the execution dependencies among components, while the structures of binomial trees facilitate the modeling of all possible execution paths.; Our architecture-based reliability model, taking the styles and grammar as inputs, not only addresses modern software infrastructures, but also resolves the traditional modeling limitations. Because of a white-box approach, this model facilitates making effective design decisions and choosing suitable software components at an early stage of the software development process. It eliminates the need for retesting the whole software system, once configurations are changed. Therefore, our architecture-based reliability model is expected to gain a much wider scope of applications. |
