Finding Upper Bounds for Software Failure Probabilities - Experiments and Results

摘要

This paper looks into some aspects of using Bayesian hypothesis testing to find upper bounds for software failure probabilities, which consider prior information regarding the software component in addition to testing. The paper shows how different choices of prior probability distributions for a software component's failure probability influence the number of tests required to obtain adequate confidence in a software component. In addition, it evaluates different choices of prior probability distributions based on their relevance in a software context. The interpretations of the different prior distributions are emphasised. As a starting point, this paper concentrates on assessment of single software components, but the proposed approach will later be extended to assess systems consisting of multiple software components. Software components include both general in-house software components, as well as pre-developed software components (e.g. COTS, SOUP, etc).