Software process research is concerned with the problem of systematically managing and automating the development and maintenance of software systems. Key challenges in software process research include understanding and describing existing processes, developing and specifying new processes, and evolving existing processes toward new processes. This project investigates techniques in two areas of software process research that fundamentally contribute to answering these challenges: (1) capture of characteristic process data and (2) analysis of process data for purposes of detailed problem identification. With such results, software engineers can gain a statistically-based understanding of an existing process, uncover systemic problems indicating areas for possible process improvement, and calculate the effects of changes to processes. The starting point for the investigation is an event-based model of software processes. An event is used to characterize the dynamic behavior of a process in terms of identifiable, instantaneous process milestones, such as invoking a development tool or deciding upon the next activity to be performed. A novel, hybrid approach integrating automated and manual data collection techniques records information about both computer-based events (e.g., tool invocation) and non-computer-based events (e.g., decision making), relating events within and across these categories. Metrics appropriate for revealing anomalous software process behavior are defined and calculated. A flexible testbed for storage and retrieval of process data with a variety of analysis tools is used as a basis for correlation analysis. Results are validated by employing the techniques in the capture and analysis of several real-world software processes used in industry.
