Are software failures chaotic?

Abstract

We report on an experimental investigation of software reliability data. Our hypothesis in this investigation is that software failures are the result of a fundamentally deterministic process, rather than being realizations of a stochastic process as is commonly assumed. Using the techniques of nonlinear time series analysis, we examine three software reliability datasets for the signatures of deterministic, and possibly chaotic, behavior. In these datasets, we have found firm evidence of deterministic behavior, and hints of chaotic behavior. However, the latter are too limited to permit a definitive conclusion about the presence or absence of chaotic behavior.