Methods and problems of software reliability estimation

There are many probabilistic and statistical approaches to modelling software reliability. Software reliability estimates are used for various purposes: during development, to make the release decision; and after the software has been taken into use, as part of system reliability estimation, as a basis of maintenance recommendations, and further improvement, or a basis of the recommendation to discontinue the use of the software. This report reviews proposed software reliability models, ways to evaluate them, and the role of software reliability estimation. Both frequentist and Bayesian approaches have been proposed. The advantage of Bayesian models is that various important but nonmeasurable factors, such as software complexity, architecture, quality of verification and validation activities, and test coverage are easily incorporated in the model. Despite their shortcomings - excessive data requirements for even modest reliability claims, difficulty of taking relevant nonmeasurable factors into account etc. - software reliability models offer a way to quantify uncertainty that helps in assessing the reliability of a software-based system, and may well provide further evidence in making reliability claims.