Predicting multi-platform release quality

Abstract

One difficulty in characterizing the quality of a major feature release is that many releases are implemented on several platforms, with each platform using a different subset of the new features. Also, these platforms can have substantially different performance expectations and results. In order to characterize the entire release adequately in predictive models, we need a robust customer experience metric that is capable of representing many disparate platforms. Several multi-platform SWDPMH (software defects per million usage hours per month) variants have been developed in an attempt to anticipate a release's overall field quality. In addition to predicting the overall release quality, it is critical that we provide guidance to business units concerning remediation of releases predicted to not achieve adequate quality, and also provide guidance regarding how to modify practices so subsequent releases achieve adequate quality. Models have been developed to both predict MP-SWDPMH and to identify specific in-process drivers that likely influence MP-SWDPMH. At this time, these modeling results can be available as early as five or six months prior to release to the customers.