Quantitative Analysis of Faults and Failures in a Complex Software System

The dearth of published empirical data on major industrial systems has been one of the reasons that software engineering has failed to establish a proper scientific basis. In this paper we hope to provide a small contribution to the body of empirical knowledge. We describe a number of results from a quantitative study of faults and failures in two releases of a major commercial system. We tested a range of basic software engineering hypotheses relating to: the Pareto principle of distribution of faults and failures; the use of early fault data to predict later fault and failure data; metrics for fault prediction; and benchmarking fault data. For example, we found strong evidence that a small number of modules contain most of the faults discovered in pre-release testing, and that a very small number of modules contain most of the faults discovered in operation. However, in neither case is this explained by the size or complexity of the modules. We found no evidence to support previous claims relating module size to fault density, nor did we find evidence that popular complexity metrics are good predictors of either fault-prone or failure-prone modules. We confirmed that the number of faults discovered in pre-release testing is an order of magnitude greater than the number discovered in 12 months of operational use. We also discovered fairly stable numbers of faults discovered at corresponding testing phases. Our most surprising and important result was strong evidence of a counter-intuitive relationship between pre and post release faults: those modules which are the most fault-prone pre-release are among the least faultprone post-release, while conversely the modules which are most fault-prone post release are among the least fault-prone pre-release. This observation has serious ramifications for the commonly used fault density measure. Not only is it misleading to use it as a surrogate quality measure, but its previous extensive use in metrics studies is shown to be flawed. Our results provide data-points in building up an empirical picture of the software development process. However, even the strong results we have observed are not generally valid as software engineering laws because they fail to take account of basic explanatory data, notably testing effort and operational usage. After all, a module which has not been tested or used will reveal no faults irrespective of its size, complexity, or any other factor.