Filtering and scalability in the ECO distributed event model

Abstract

Event-based communication is useful in many application domains, ranging from small, centralised applications to large, distributed systems. Many different event models have been developed to address the requirements of different application domains. One such model is the ECO model which was designed to support distributed virtual world applications. Like many other event models, ECO has event filtering capabilities meant to improve scalability by decreasing network traffic in a distributed implementation. Our recent work in event-based systems has included building a fully distributed version of the ECO model, including event filtering capabilities. This paper describes the results of our evaluation of filters as a means of achieving increased scalability in the ECO model. The evaluation is empirical and real data gathered from an actual event-based system is used. The findings show filters to be highly valuable in making distributed implementations of the model scale, that multicast contributes to the scalability and, perhaps most signifcantly, that multicast groups can be dynamically generated from filters using local (per node) rather global knowledge of the distributed application.