Optimal Software Testing across Version Releases
Simon Wilson, Seán Ó Ríordáin · 2018
Much software is now updated with a regular schedule of version releases. This chapter looks at models for bug detection in software where this is the case. It illustrates the ideas through a well-known non-homogeneous Poisson process software reliability model due to Goel and Okumoto. Model inference is developed from a Bayesian perspective and applied to data on bug reports for several versions of Mozilla Firefox. The model is applied to the decision problem of determining the optimal time between releases based on bug detection data. An application of this fitted model, to the question of how much time there should be between releases, has also been demonstrated for the case where this time is fixed across all releases with a decision theoretic solution. The chapter discusses several approaches that generalize existing software reliability models to the case of a sequence of releases.