Event-Based Multi-level Service Monitoring

Due to the growing pervasiveness of the service paradigm, modern systems are now often built as software as a service, and tend to exploit underlying platforms and virtualized resources also provided as services. Managing such systems requires that we be aware of the behaviors of all the different layers, and of the strong dependencies that exist between them. In this paper we present the Multi-layer Collection and Constraint Language (mlCCL). It allows us to define how to collect, aggregate, and analyze runtime data in a multi-layered system. We also present ECoWare, a framework for event correlation and aggregation that supports mlCCL, and provides a dashboard for on-line and off-line drill-down analyses of collected data. Empirical assessment shows that the impact of the approach on runtime performance is negligible.