Performance Evaluation of Event-Driven Software Applied in Monitoring Systems

Real-time monitoring systems now often use complex event processing software and in this paper we analyze performance of such an approach in a simple case study. After introducing briefly the main concepts of event processing networks we define the case of detecting anomalies in a stream of two events which, in correct operation, should appear in pairs keeping precise temporal relation. For such a stream we propose two event processing networks for detection of specific anomalies which are then implemented on the PROTON software platform and tested as RESTful web applications. The performance tests include two scenarios for generation of the input events and show the average system response time as a function of the increasing load (the stream of incoming events). The results identify limitations of the tested implementation which are related to deficiencies of massive multi-threaded processing in the Java environment.