A Model for Failures in Dependable Systems

This document describes a model for failures in dependable systems. A general failure model is described in terms of a system consisting of interacting components. This model is then applied to an object-based interaction model. The model is based on events which occur with some value at some time. Components in the system observe events and have expectations which define regions in a value, time space. A failure is detected when what is observed does not match what is expected. The concepts in the model can be used to analyse a given configuration of engineering mechanisms, application components and infrastructure to determine what failures can and cannot be tolerated by this configuration. This can then be mapped onto an application-level statement: what failures the applications can and cannot tolerate. The intention is that the model should provide the underlying framework for further work on dependable distributed computing. Some familiarity with basic principles of object-based distributed computing is assumed.