Software is a ubiquitous feature of today's world. The goodness of products and services is frequently dependent on the goodness of the related software. Optimal software performs correctly and requires minimal effort and cost to develop and maintain. The development of optimal software is an admirable goal but is difficult to achieve. In particular, software maintenance and evolution is costly and error-prone. The significance of the problem is magnified for high-assurance applications that require the certainty that the software will behave reliably despite budget constraints and product evolution. Though automated software development is the ideal solution, design for evolution is the practical solution. For most applications, analysis of the required behavior (behavioral analysis) and translation into a blueprint for building the software (software design) are necessary. High-level design involves the organization of the required behavior into building blocks or components. Design for evolution is the generation of a software architecture that can be changed with minimal human effort to produce a class of similar applications. Design for evolution makes feasible the cost-effective development of high-assurance applications.

This dissertation presents a semi-automatable research approach for designing an evolvable software architecture. The research approach focuses on the partition of basic elements of a software solution into reusable components that localize the effects of change. The input to the partitioning process is a set of software requirements along with an analysis of the required behavior and planned or feasible evolution of the product line. The output is a partition of the required behavior into components that reduce the effort associated with developing a software product line. The dissertation provides an analytical verification of the research approach through proof and constructive examples. Empirical results validate the effectiveness of the research approach in comparison to human intuition, experience, or other training. The research approach is novel and fills a gap in the systematic generation of software architectures that minimize the effort associated with product-line evolution. The dissertation describes in detail the degree to which the research approach is automatable and specifies, more generally, future research needed to achieve full automation of software architecture generation.

350 pages

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu