TSpace

Abstract (summary): Building systems that satisfy the given requirements is a main goal of software engineering. The success of this process relies largely on the presence of an adequate architectural design. Traditional paradigms deal with requirements separately from design. Our empirical studies show that crossing the boundary between requirements and design is difficult, existing tools and methods for bridging the gap inadequate, and that software architects rely heavily on experience, prior solutions, and creativity. Current approaches in moving from requirements to design follow two schools. One is architecture-centric, focused on providing assistance to architects in reuse. The other is requirements-centric, and tends to extend established development frameworks and employ mappings to transition from requirements to architecture. Jackson indicates that clear understanding of requirements (the problem) is crucial to building useful systems, and that to evolve successfully, their design must reflect problem structure. Taylor et al. argue that design is the central activity in connecting requirements and architecture. Nonetheless, existing approaches either overlook underlying structure of requirements or design considerations. This dissertation presents a novel theory enabling requirements structuring and design analysis through requirements refinement and refactoring. The theory introduces a refinement process model operating on four abstraction levels, and a set of refactoring operators and algorithms. The method works in small, well-guided steps with visible progress. The theory provides a basis for designers to analyze and simplify requirement descriptions, remove redundancy, uncover dependencies, extract lower-level requirements, incorporate design concerns, and produce a system decomposition reflecting the underlying problem structure. A design built on top of this decomposition is better suited for evolution than one created without explicit structural analysis. The theory is validated on an industrial-sized project, wherein a suitable system decomposition is produced and a comparison made to a conventionally-devised solution. Examples demonstrate that the theory handles changes incrementally. It is explained how the theory addresses the existing challenges in going from requirements to design and supports fundamental software engineering principles. The method is assessed against common validation criteria. The approach is compared with prominent related work.