Simple Metrics for Improving Software Process Performance and Capability: A Case Study

Abstract

This paper presents a case study which describes the application of statistical process performance monitoring and capability assessment on a large process control software project. The process performance and capability monitoring used in the case study is part of a Quality Framework for Software Development (QFSD) devised for the development of control systems at Fisher–Rosemount Systems and provides a practical alternative to ISO9000 quality models. The process performance measurements were based on commonly available metrics that could be obtained with the minimum disruption to the processes being examined. The application of statistical methods were used to establish problem areas at the earliest opportunity allowing process adjustments to be made to improve the process performance. The processes used were defined in terms of work products which were categorized and evaluated for the level of completion. This allowed a process capability to be calculated. Those processes which were found to be at a lower capability level became the focus for the process improvement for the next project. In particular the capability measurement identified processes where questions needed to be raised about the relevance of the process, whether the techniques used were effective and whether the tools were adequate. It was found that achieving high performance and capability is a learning process with the development team improving with each new project. The benefits have been better managed, more cost and time effective projects producing higher quality software. Management and the development team have obtained a better understanding of the software development process and this continuing learning process has lead to a continuous improvement in both the development methodology and the resulting software. The paper presents examples of the real benefits that can be obtained by tracking process performance and assessing process capability at all stages, and shows that this can be achieved without the need to resort to complex procedures for process measurement, as most of the empirical data came from commonly available process data.