Combining various solution techniques for dynamic fault tree analysis of computer systems

Abstract

Fault trees provide a conceptually simple modeling framework to represent system-level reliability in terms of interactions between component reliabilities. DIFtree (Dynamic Innovative Fault trees) effectively combines the best static fault tree solution technique (binary decision diagrams) with Markov solution techniques for dynamic fault trees. DIFtree includes advanced techniques for modeling coverage; coverage modeling has been shown to be critical to the analysis of fault-tolerant computer systems. DIFtree is based on a divide-and-conquer technique for modularizing the system-level fault tree into independent sub-trees; different solution techniques can be used for sub-trees. In this paper, we extend the DIFtree analysis capability to model several different distributions of time-to-failure, including fixed probabilities (no time component), exponential (constant hazard rate), Weibull (time-varying hazard rate) and log-normal. Our approach extends both the BDD and Markov analytical approaches and incorporates simulation as well.