Donald Bren School of Information and Computer Sciences

One central feature of the structure of a software system is the coupling among its components (e.g., subsystems. modules) and the cohesion within them. The purpose of this study is to quantify ratios of coupling and cohesion and use them in the generation of hierarchical system descriptions. The ability of the hierarchical descriptions to localize errors by identifying error-prone system structure is evaluated using actual error data. Measures of data interaction, called data bindings, are used as the basis for calculating software coupling and cohesion. A 135,000 source line system from a production environment has been selected for empirical analysis. Software error data was collected from high-level system design through system test and from some field operation of the system. A set of five tools is applied to calculate the data bindings automatically, and cluster analysis is used to determine a hierarchical description of each of the system's 77 subsystems. An analysis of variance model is used to characterize subsystems and individual routines that had either many/few errors or high/low error correction effort.