A discrete software reliability growth model with two types of errors

Abstract

AbstractAs the social missions of computer systems increase, it becomes more important to develop high‐reliability software. For this reason software systems are tested repeatedly to remove latent errors in the testing phase during software development. One of the most interesting methods is to evaluate software reliability by using test data observed in the error‐detection process. In this paper we model the error‐detection process as a stochastic model and introduce several effective measures to evaluate software reliability. In modeling we use the number of test run trials as a unit of the error‐detection period and assume that the cumulative number of software errors detected in an arbitrary testing time interval is a nonhomogeneous Poisson process. Further, we consider the difficulty of error detection because the reliability of software is evaluated from the characteristics and frequency of the errors. Applying the model to the actual test data, we perform the goodness‐of‐fit test and infer the software reliability measures. Finally, we discuss an optimum software release problem based on the software reliability index.