Estimation of coverage probabilities for dependability validation of fault-tolerant computing systems

Abstract

Dependability validation is a major step toward development of high-assurance computing systems. This paper addresses the problem of estimating the coverage probabilities by statistically processing the information collected through physical or simulated fault injection. 3-stage random sampling is employed to derive the means, variances and confidence intervals of the coverage probabilities. The statistical experiments are carried out in a 3D fault space that accounts for system inputs, fault injection times and fault locations. In the case of real-time systems, the inputs and the injection times also provide useful information about the workload to be executed. The proposed solution technique is tested against the data generated by a program that mimics a fault environment. Two application examples are considered. Several working rules for designing 3-stage random sampling experiments are also provided.<>