Improving Service Diagnosis through Increased Monitoring Granularity

Abstract

Due to their loose coupling and highly dynamic nature, service-oriented systems offer many benefits for realizing fault tolerance and supporting trustworthy computing. They enable automatic system reconfiguration in case that a faulty service is detected. Spectrum-based fault localization (SFL) is a statistics-based diagnosis technique that can effectively be applied to pinpoint problematic services. It works by monitoring service usage in system transactions and comparing service coverage with pass/fail observations. SFL exhibits poor performance in diagnosing faulty services in cases when services are tightly coupled. In this paper, we study how and to which extent an increase in monitoring granularity can help to improve correct diagnosis of tightly coupled faulty services. We apply SFL in a real service-based system, for which we show that 100% correct identification of faulty services can be achieved through an increase in the monitoring granularity.

Analysis of service diagnosis improvement through increased monitoring granularity

“…7.2. The overhead results presented in Table 9 are different from the results obtained in our previous overhead experiments outlined in our earlier article (Chen et al 2013). In this other article, the experiments were only aimed at getting an initial feeling of the potential overhead caused by various monitoring strategies, and we had to circumvent a few flaws in the implementation.…”

“…The Turmeric monitor that we implemented for the experiments in (Chen et al 2013) caused a large amount of overhead. The major reason for this overhead was due to the use of synchronous database access to record the monitoring data.…”

“…This current paper is an extension of our previous work presented in (Chen et al 2013). The previous article is focused on the improvement of the diagnosis through increasing the monitoring granularity with a preliminary overhead assessment.…”