ANALYZING ERROR-PRONE SYSTEM STRUCTURE

One central feature of the structure of a software system is the nature of the interconnections among its components (e.g., subsystems, modules). The concepts of coupling and strength have been used in the past to refer to the degree of interconnection among and within components. The purpose of this study is to quantify ratios of coupling and strength and use them to identify error-prone system structure. We use measures of data interaction, called data bindings, as the basis for calculating software coupling and strength and analyzing system structure. We selected a 148 000 source line system from a production environment for empirical analysis. We collected software error data from high-level system design through system test and from some field operation of the system. We describe the methods used for gathering data during the ongoing project and characterize the software error data collected. We apply a set of five tools to calculate the data bindings automatically and use a clustering technique to determine a hierarchical description of each of the system's 77 subsystems. A nonparametric analysis of variance model is used to characterize subsystems and individual routines that had either many or few errors or high or low error correction effort. The empirical results support the effectiveness of the data bindings clustering approach for localizing error-prone system structure. Routines with the highest coupling/strength ratios had 7.0 times more errors per KNCSS (1000 source statements excluding comments) than did routines with the lowest coupling/strength ratios. Subsystems with high coupling/strength ratios had routines with 4.8 times more errors per KNCSS than did subsystems with low coupling/strength ratios. The interpretations span several areas: development methodology, inspection methodology, data collection and analysis, size, coupling/strength, and system structure.