Data-driven software reliability and availability modeling and prediction

Traditional software deployment readiness criteria, such as ‘zero severity one defects’, do not provide any indication of how reliable the product will be in the field. In this paper, we propose a software reliability prediction framework to achieve data-driven, customer focused reliability and availability assessment throughout the entire development life cycle. Focusing on front-end reliability and availability improvement, the framework starts with availability evaluations as early as the architecture design phases. Markov-based architecture reliability models are used to study the failure and failure recovery mechanisms of the systems and solutions. These early evaluations can help architecture design, reliability requirement setting and reliability budget allocation. The early phase models and predictions can be updated as testing data becomes available. Software reliability growth models (SRGMs) are used to estimate one of the most influential parameters, i.e. the failure rates of software. Estimation of other reliability parameters, such as coverage factor, silent failure detection times and recovery durations and success probabilities are also discussed in this paper. This framework also calibrates test data with field observations, and thus forms a close-loop approach to evaluate the reliability and availability of the software product to verify that the product meet specific reliability expectation.