Probabilistic Treatment of Service Assurance in Distributed Information Systems

The paper describes a probabilistic view of QoS (quality of service) assurances provided by a distributed information system (DIS). The service specs factor in the inaccuracies and partial knowledge of system models pertaining to failures of infrastructure components. These uncertainties have a compounded effect on the robustness of applications running on a DIS. We associate quantitative properties to the objects managed by a middleware service-layer in terms of probabilistic assertions: such as stability and boundedness in object behaviors. The properties are derived by analyzing the component-level behaviors composed into a service-layer object. For e. g., the fault-tolerance of a replicated data system is determined from individual data server failure probabilities. Such metrics enable accurate decision-making by the application-layer control algorithms when responding to fault events. The paper describes a case study of web-based information services, using server replication mechanisms.