Online Reliability Time Series Prediction for Service-Oriented System of Systems

A Service-Oriented System of System (or SoS) considers system as a service and constructs a value-added SoS by outsourcing external systems through service composition. To cope with the dynamic and uncertain running environment and assure the overall Quality of Service (or QoS), online reliability prediction for SoS arises as a grand challenge in SoS research. Online reliability prediction is expected to pave the way to deploy fault removal and self-(configuration, healing, optimization, or protection) (or self-*) SoS that can automatically adapt to the complex and dynamic running environment. In this paper, we propose a novel approach for component level online time series reliability prediction based on Probabilistic Graphical Models (or PGMs). We assess the proposed approach via invocation records collected from widely used real web services and experiment results demonstrate the effectiveness of our approach.