Causal Inference Based Service Dependency Graph for Statistical Service Fault Localization

Abstract

In the interconnection environment, people combine basic services into composite services to provide more complex function for sophisticated applications. Accordingly, service fault localization in composite services becomes a critical issue for guaranteeing the normal running of composite services. This paper proposes a novel Causal Inference based Service Dependency Graph (CISDG) for statistical service fault localization. Our approach first utilizes the dependencies between basic services in the composite services by transforming the service dependency graph into a causal graph. Then it intuitively applies the causal inference to service fault localization on the composite services. Our work mainly focuses on developing CISDG and the causal inference on CISDG. To develop CISDG, we characterize the dependency and causal relationships between basic services and the target causal graph. To perform the causal inference on CISDG, we apply the well-known causal inference techniques such as the Back-Door Criterion and enhance the algorithm of the network diagnostic algorithm based Causal Inference (CIND) to improve the efficiency of the statistical service fault localization. The case study illustrates that our approach has advantages over its rivals in the service fault localization of composite services.