Parameter Dependencies for Component Reliability Specifications

Predicting the reliability of a software system at an architectural level during early design stages can help to make systems more dependable and avoid costs for fixing the implementation. Existing reliability prediction methods for component-based systems use Markov models and assume that the software architect can provide the transition probabilities between individual components. This is however not possible if the components are black boxes, only at the design stage, or not available for testing. We propose a new modelling formalism that includes parameter dependencies into software component reliability specifications. It allows the software architect to only model a system-level usage profile (i.e., parameter values and call frequencies), which a tool then propagates to individual components to determine the transition probabilities of the Markov model. We demonstrate the applicability of our approach by modelling the reliability of a retail management system and conduct reliability predictions.