Software Metrics: Don't just track - know when to act!

摘要

Software metrics are an essential tool for project and quality management, but do you know when your metrics are telling you to take action? Can you identify significant trends and process changes or deviations? Do you know if process shifts are statistically significant? Tracking process and product metrics gives you insight into your development process. Comparisons to historical or industry data or well-defined targets bring you closer to understanding if current data is good news or cause for alarm. Even better are statistical limits which help to take the guess-work out of metrics analysis and empower software developers, project managers, and quality engineers to knowledgeably control development processes and achieve product quality objectives. Statistical Process Control (SPC) techniques, such as process control charts, are common in manufacturing environments, but uncommon in the software industry due to many implementation problems inherent to the nature of software development. However, as in manufacturing, SPC techniques are extremely useful for understanding process capabilities, quickly identifying process deviations, and controlling product quality. In this paper, some of the challenges of implementing SPC for software processes are discussed, several methods for addressing the problems unique to SPC use within software development are described, and practical examples of SPC implementation across the software development lifecycle are explained. Other traditional quality tools, such as Pareto analysis, are used to augment metrics analysis. Real-life examples from large and complex industry projects developed at IBM~R Rochester are used to illustrate SPC use in development as well as the maintenance phase of the software lifecycle, while the caveats/limitations of applying control charts in software environments are addressed. Examples of the analysis of process outliers and actual implemented actions are also provided. The project examples given are based mainly on releases of the operating system of the IBM eServer iSeries.