Towards Online Localization and Recovery for Faulty Components in Component-Based Applications

Off-The-Shelf (COTS) software components have been extensively used by applications over the world. However, COTS components always carry issues that might and sometimes only take place at runtime, in particular, when combined with other components. This might bring serious trouble for the application´s dependability. Based on the industry experiences that lots of faulty components usually occupy excessive resources, this paper proposes an approach to localize the faulty components in an application automatically through analyzing the resource usage of components. Once a faulty component interferes with the application´s performance, our approach generates an anomaly report, localizes the faulty component, and enables component level recovery to remove the negative impact. We have implemented a prototype and demonstrated its effectiveness on the well-known JPetStore benchmark with performance overhead of less than 3%.