Determining the expected time to unsafe failure

The number of applications requiring highly reliable and/or safety-critical computing is increasing. One emerging safety metric is the Mean Time To Unsafe Failure (MTTUF). This paper summarizes a novel technique for determining the MTTUF for a given architecture. The first step in determining the MTTUF for a system is to estimate system Mean Time To Failure (MTTF) and system fault coverage. Once these two parameters are known then the system MTTUF can be calculated. The presented technique allows MTTF and system coverage to be estimated from dependability models that incorporate time varying failure and/or repair rates. Existing techniques for the estimation of MTTUF require constant rate dependability models. For the sake of simplicity, this paper uses Markov models to calculate MTTUF. The presented approach greatly simplifies the calculation of system MTTUF. Finally a comparison is made between reliability expected time metrics (MTTF and MTBF) and safety expected time metrics (MTTUF and MTBUF).