Toward a New Theory of Software Defects: Leveraging Software Variance and Heterogeneity To Improve Process Efficiency

Contrary to many other industrial processes, software production is characterized by an unusually high variance. This directly results from the significant role of the human factor in all its phases. As a result, measurements and modeling, which are predicated on the homogeneous nature of software, do not provide dependable means of process control and analysis. This paper first analyzes various aspects of software metrics that demonstrate the heterogeneous nature of software and the variance in the software production process. Then, a new model of software is developed to better represent software behavior. The approach is based on the empirical evidence that software defects tend to congregate in “design holes ” which are deficiencies analogous to poles in linear systems. These are software instabilities. The techniques utilizes the “fingerprints ” left by software changes on the software map. Two major applications are derived from the model analysis. First, the model provides a method for evaluating the “goodness ” of the software architecture. The model can also be used to balance the budgeting of the testing effort among the various software functionalities and their interactions. 1