Online Anomaly Detection for Service-Oriented Components in OSGi-based Applications

OSGi has become one of the most promising frameworks for managing service-oriented and component-based applications. The OSGi-based service-oriented components delivered by different vendors are usually black-box program units which lack source code and design documents. Thus, it is difficult to evaluate their quality by static code analysis, and the defective components may lead to the failure of the whole system eventually. In this paper, we propose an online method for detecting anomalous service-oriented components in OSGi-based applications. A thread-tracing method is presented to monitor resource utilization and interactions between components. The method considers both the dynamic service invocation and static method invocation. Furthermore, according to the monitored data, we detect anomalous components by control charts, which can detect the anomalous trend of resource utilization without prior knowledge. A prototype tool was implemented and applied to a real application server. The experimental results show that our method 1) is of high accuracy for monitoring resource utilization in component perspective; 2) does not introduce significant overhead; 3) and can detect anomalous components effectively.