Abstract:

   Current software reliability evaluation methods do not take into account the execution probability of each function and the degree of failure severity. We propose a new method for evaluating software reliability based on the software architecture. The UML-described software architecture is analyzed. Execution probability of the functions is derived, and reliability of components and connectors defined. The degree of failure severity of components and connectors are obtained according to expert experiences. A Markov model is constructed from various scenarios, and reliability of individual components and the overall system estimated using the model.  Validity of the method is shown with an ATM example.