Fault Tolerant Systems Analysis: Dynamic Combinatorial Models.

摘要

The problems that arise during reliability analysis of a fault tolerant computer system can be broadly classified into those relating to the construction of the model, and those relating to the solution of the model. The construction of a model of a complex fault tolerant system consists of selecting an appropriate 'language' for the description of the system, abstracting the important characteristics of the system to be studied, and expressing these characteristics in the description language. The underlying stochastic representation of the system can then be automatically determined from the description language; the solution of the underlying stochastic process provides estimates of the desired measures. Some examples of modeling languages that are appropriate for simplifying the model construction task are combinatorial models, such as reliability block diagrams (20) and fault trees (2). Such combinatorial models are useful because they provide a concise representation of the system; however, they are not able to model the dynamic system behavior in response to a fault or an error. The first topic considered under the auspices of this grant was concerned with the development of techniques for incorporating fault and error modeling techniques into combinatorial models. A second area of research conducted under the current contract concerns the development of fast, accurate algorithms for the solution of fault tree models. Several different techniques were developed for producing bounded approximations for both static and dynamic combinatorial models. (The techniques were applied specifically to fault trees, but are also applicable to reliability block diagrams.) Techniques for the consideration of truncated fault trees were derived which could be used to produce bounded estimates of system reliability from partially developed fault trees.