Measurement processes are software, too

Abstract

Software process improvement and measurement are closely linked: measures are the only way to prove improvements in a process. Despite this link, and the interest in process improvement, measurement is not widely applied in industrial software production. This paper describes a method designed to guide the definition, implementation and operation of measurement processes. The method, which builds upon Fenton's measurement framework and GQM, starts from the point that measuring a software process is in its turn a process in the software process. The three basic ideas of the method are derived from this assumption: the measurement process should reuse and suitably adapt the same phases of the software process: requirements definition, design, implementation, etc. A descriptive process model should be the essential starting point of a measurement process. Many concepts and tools which derive from the object oriented approach should be effectively used in the measurement process. An experimental application in an industrial process has shown that building the process model was the hardest part of the measurement process, and that it has improved the quality of measurement by reducing misunderstandings. Object oriented concepts and tools make it possible to automate certain tasks (for instance the definition of the schema of the measurement database) and to improve robustness against changes in the measurement process.