PublicaHier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.
Measuring Software Inspection Processes
Software inspections are accepted as a formal, effective, and efficient method for defect detection that is applicable in all development phases. In order to control the performance of software inspection processes the collection and analysis of data is recommended in the literature. Suggestions have been made for measuring selected aspects of inspections. Yet, we found no comprehensive survey of measuring software inspections. In this thesis, we propose a five-level measurement framework that includes most applications of measurement in the context of software inspections. Quantitative models published in the literature are collected and discussed according to the framework's five level structure. This collection of models is intended to be a comprehensive survey of the topic. Moreover, the framework can be applied in future projects to derive a set of models that directly serve the intended measurement purpose. The survey shows that inspection efficiency is considered animportant co mponent of inspection measurement. However, most published efficiency models for inspections have shortcomings. To address these shortcomings, we developed a new efficiency model that characterizes the efficiency of a single inspection. We performed simulations to analyse and assess the new efficiency model. Data published by the software engineering community was collected for the model's input variables. Thus, our simulations are based on real data and the efficiency results of the simulations mirror typical inspection efficiency. Therefore, the results can be used for benchmarking purposes. Important results of the simulations are 1) that software inspections cause a reduction of software development cost, 2) that design inspection efficiency is higher than code inspection efficiency, 3) that inspection efficiency should be evaluated in the context of other defect detection techniques applied during development, and 4) that the average cost per defect for inspectionsexplains most o f the model's variation in inspection efficiency.