SFDCloud: top-k service faults diagnosis in cloud computing

Abstract

With a variety of providers large and small delivering a number of cloud-based services, cloud computing is evolving into an important service delivery infrastructure. One of the challenges in this evolution is how to provide necessary fault handling for migration long-running or computationally-intensive application services into shared open cloud infrastructures. To minimize failure impact on services and application executions, we present a diagnostic architecture and a diagnosis method based on the service dependence graph (SDG) model and the service execution log for handling service faults. By decoupling diagnosis service components and sharing diagnosis resources, the scalability of diagnosis methods is improved by incorporating third-party diagnostic components into our architecture. By analyzing the dependence relations of activities in SDG model, our diagnosis method identifies the incorrect activities, and explains the root causes for the web service composition faults, based on the differences between successful and failed executions of composite service. Experimental results show that our method is effective in diagnosing faults in web service composition of various scales.