Configurable Fault Trees

摘要

Fault tree analysis, as many other dependability evaluation techniques, relies on given knowledge about the system architecture and its configuration. This works sufficiently for a fixed system setup, but becomes difficult with resilient hardware and software that is supposed to be flexible in its runtime configuration. The resulting uncertainty about the system structure is typically handled by creating multiple dependability models for each of the potential setups. In this paper, we discuss a formal definition of the configurable fault tree concept. It allows to express configuration-dependent variation points, so that multiple classical fault trees are combined into one representation. Analysis tools and algorithms can include such configuration properties in their cost and probability evaluation. The applicability of the formalism is demonstrated with a complex real-world server system.