Probabilistic modeling of hardware and software interactions for system reliability assessment

An important aspect of complex hardware-software systems reliability is the interaction between hardware and software. Most of the existing work has assumed either independence between hardware (HW) and software (SW) or a fixed proportion of hardware reported failures to represent HW/SW interactions. These assumptions do not necessarily reflect reality. In this paper, probabilistic HW/SW interactions, in conjunction with hardware and software reliability, are considered to model and assessment overall hardware-software systems reliability. However, incorporating uncertainty demands the use of stochastic programming to estimate the parameters of the hardware reliability model. To capture the HW/SW interactions, a Markov Stochastic process model with uncertainty in transition rates is proposed and solved using Monte Carlo sampling. By combining hardware, software, and probabilistic HW/SW interactions, the overall system reliability can be predicted not only as a point estimator but as quantiles or confidence intervals. The proposed methodology was demonstrated with data from a real computer system provided by Los Alamos National Laboratory. Reliability predictions for a 1,095-day horizon were carried out with an R script in approximately 1.5 s running on a laptop, which demonstrates the feasibility and convenience of the approach. In the four cases analyzed, the 97.5% lower bound of system reliability estimation was above the reliability calculated assuming independent hardware and software reliability.