Improving Service Diagnosis through Invocation Monitoring

Service oriented architectures support software runtime evolution through reconfiguration of misbehaving services. Reconfiguration requires that the faulty services can be identified correctly. Spectrum-based fault localization is an automated diagnosis technique that can be applied to faulty service detection. It is based on monitoring service involvement in passed and failed system transactions. Monitoring only the involvement of services sometimes leads to inconclusive diagnoses. In this paper, we propose to extend monitoring to include also the invocation links between the services. We show through simulations and a case study with a real system under which circumstances service monitoring alone inhibits the correct detection of a faulty service, and how and to which extent the inclusion of invocation monitoring can lead to improved service diagnosis.