A Semi-Parametric Approach to Testing for Reliability Growth, with Application to Software Systems

We consider the following general model for reliability growth: the distribution of times between failures belongs to a known parametric family (not necessarily exponential), and the parameter corresponding to the distribution of a particular time between failures is either an unknown constant or an unobservable random variable with a (possibly unknown) distribution which can depend on past observations. We propose that acceptable reliability can sometimes be formalized as a state in which the value of the parameters is lower than a level set before testing begins. We apply sequential detection methodology to the problem of ascertaining that an acceptable state of reliability has been attained and illustrate our approach by applying it to testing for reliability growth of a software system, using actual data.