A requirements-based software reliability model

Experts in the field of software reliability modeling have several issues with the current state of software reliability models. Software reliability models that make the assumption that past failure history data is the best indicator of future behavior still dominate the discipline. Because of this assumption, software reliability models treat the process and the product as a black box and therefore can not explicitly be used for assessing the software process or product. Software reliability models that currently predict reliability early in the life cycle lack a sound statistical foundation. Finally, software reliability models should contribute to understanding and removing faults to improve reliability.; The thesis of this dissertation is that it is possible to address these issues using a "systems engineering" approach to software reliability assessment and prediction. A requirements based software reliability modeling approach provides the structure and organization to integrate several engineering concepts into software reliability modeling. This model shifts the focus of software reliability modeling from modeling failure history to focusing upon the requirements. Reliability assessment is based upon validation of requirements, root cause analysis of their failure modes, and structural analysis of the functional architecture of the software. Failure modes and fault trees are used to develop a component based reliability model. The component based approach to reliability assessment allows for early assessment of reliability. Bayesian inference provides the statistical basis for quantifying the software reliability at the component level. A framework for the model is developed and illustrated through a simple experiment. The reliability data sources are explored and means of incorporating different sources of data examined. The dissertation demonstrates how the component nature of the model allows the model to adjust its assessment of reliability based upon the changing requirements, without loss of significant reliability data.