Efficient Sensitivity Analysis of Reliability in Embedded Software

The reliability of software architectures is an important quality attribute to consider in the design of embedded software systems. Reliability is especially important in safety-critical systems where human health and safety is often at stake. Researchers have developed a number of models that can estimate the reliability of software systems at design time. These models use matrix operations of considerable complexity, such as matrix inversions. During the design process of an embedded system as used by the automotive industry, the reliability of a system has to be evaluated after each change in the design. Such safety-critical systems are often subjected to sensitivity analysis, where a single parameter is changed numerous times to estimate its effect on the safety of a system, with a subsequent re-evaluation of the system’s reliability. In this paper, we introduce an efficient sensitivity analysis technique which does not require the complete re-evaluation of the system reliability. The approach uses the ∆ Evaluation technique, which computes the change in reliability of the part of the system architecture which was affected by a change. Results from experiments based on a real casestudy from the automotive industry indicate a significant improvement in time of the proposed approach compared to traditional sensitivity analysis methods.