Modeling the Impact of Change on Software Intensive Systems

Abstract

Abstract

Most significant software-intensive systems undergo substantive change/evolution during their life time of service. Managing the consequent software changes is a difficult and costly task. In this thesis, we use two different approaches to investigate system change and its impact on the architecture and design of the system. The first approach involves traditional software change impact analysis. We propose a new and different traceability model, which is based on Genetic Software Engineering (GSE). The proposed traceability model exploits some features of GSE to create a number of advanced properties that are rare in other traceability models. For example, once a software change has been fully captured, some other design documents including the component architecture and component behavior can be automatically generated/updated. All the consequent change impacts are presented in a clear way. We have also introduced the concept of evolutionary design documents that show the evolution process of a system’s architecture as well as the design of individual components. Using this proposed traceability model, a practical method to normalize and simplify the component architecture of software intensive systems has been developed. An important result we have proved is that the component architecture of a software system is independent to the functional requirements of the system. We claim that a normalized software system is easier to maintain and change. The second approach starts from a macro view. Rather than exploring the details of the change impacts from individual changes, this approach focuses on the common properties of the architecture evolution of complex systems; it stresses the topological structure from an evolutionary viewpoint. For this investigation we use scale-free networks and hierarchy theory as the major tools. Hierarchy is a natural structure for diverse large and complex systems, and recent studies reveal that many large networks from different domains are scale-free. In this research, we have discovered that the component dependency networks of many software systems are scale-free; we have also found that there is a close connection between the scale-free feature and the optimization of sorting algorithms. These results imply that there are fundamental rules working behind the evolution of large systems including software intensive systems, and that the scale-free property can be used as a possible index for the optimization level of the structure of a system. Software change and software evolution are critical aspects of software engineering. This thesis has used a macroscopic and technical, formal approach to make positive contributions to understanding and accommodating change of software-intensive systems.