This paper presents the concepts and philosophy behind a new program for the modeling and evaluation of fault-tolerant systems with extremely high reliability--HARP (Hybrid Automated Reliability Predictor). The fundamental approach of HARP is based on: (1) a reduction of state space by the behavioral decomposition of a reliability model along temporal lines into fault-occurrence and fault-handling submodels, and (2) the solution of these submodels using appropriate analytic or simulative techniques.
While this philosophy is not new and has been employed before, what is distinct is, as the authors have described in their Conclusions: (1) flexibility in the specification of fault-handling behavior through the use of an Extended Stochastic Petri Net, (2) new methods of aggregating fault-handling and fault-occurrence results, and (3) provision of automatic sensitivity analysis of system reliability with respect to parametric and initial state errors.
Thus, this is an elegant solution to the reliability prediction problem, and it removes many serious limitations of previous models. In particular, the capability of performing sensitivity analysis must be regarded as an important contribution.
Besides its technical merits, I also found the paper highly readable. Concepts are carefully presented, and references are adequately cited. Furthermore, it contains quite a bit of background and tutorial materials. This paper, together with an earlier one by two of the present authors [1], should constitute a comprehensive review and update of the state-of-the-art in reliability prediction.