A Time/Structure Based Model for Estimating Software Reliability

Estimation of the probability of software failure during a speciied exposure period, also known as reliability, has long been an important subject of research. Several models for estimating software reliability exist. The Goel-Okumoto model and the Musa execution time model are two well known time-based models. Common characteristics of time-based models include neglect of program structure and use of time dependent data in the estimation of software reliability. In this paper we point out some fundamental problems with such an approach. We proposed two distinct approaches to reliability modeling. The approach described here leads to a class of models known as time/structure based models which extend the existing time-based models by using the notion of useless testing eeort. The other approach, described elsewhere, is purely structure based and makes use of the notion of code coverage and other fault related parameters. A common characteristic of both the approaches is that they incorporate features of testing e.g. data ow or mutation testing, during the integration test phase of the software development cycle. While the pure structure based approach requires the use of certain testing methods, the time/structure based approach does not. We expect these approaches to lead to new and novel ways of estimating software reliability.