Multi-dimensional modelling and measurement of software designs

• Authors: Martin Shepperd;Darrel Ince
• Venue: CSC '90 Proceedings of the 1990 ACM annual conference on Cooperation
• Year: 1990

Design structure measures are example of a class of metrics that may be derived early on in a software project; they are useful indicators of design weaknesses - weaknesses which, if uncorrected, lead to problems of implementation, reliability and maintainability. Unfortunately, structure metrics are limited in their ability to model system architecture since they are insensitive to component size. Thus, architectures that trade structural complexity for very large components may not be detected. This paper has two concerns. First, we consider the problem of adequately measuring component size at design time. Various existing metrics are evaluated and found to be deficient. Consequently, a new, more flexible approach based upon the traceability from system requirements to design components, is proposed. Second, we address the issue of multi-dimensional modelling (in this case structure and size). We apply outlier analysis techniques to identify three classes of problem design component and relate our work to an empirical study of 62 modules. The results suggest that augmenting the more traditional approach of a single structure metric with an additional perspective, that of module size, considerably enhances the ability of design metrics to isolate problem components. It is our belief that far more sophisticated software modelling techniques, such as the multi-dimensional approach we present, are required, if measurement and modelling is to reach its full potential, as an integral part of software engineering processes.