Providing probabilistic latency bounds for dynamic publish/subscribe systems

Abstract. In the context of large decentralized many-to-many communication systems it is impractical to provide realistic and hard bounds for certain QoS metrics including latency bounds. Nevertheless, many applications can yield bet-ter performance if such bounds hold with a given probability. In this paper we show how probabilistic latency bounds can be applied in the context of pub-lish/subscribe. We present an algorithm for maintaining individual probabilis-tic latency bounds in a highly dynamic environment for a large number of sub-scribers. The algorithm consists of an adaptive dissemination algorithm as well as a cluster partitioning scheme. Together they ensure i) adaptation to the individual latency requirements of subscribers under dynamically changing system proper-ties, and ii) scalability by determining appropriate clusters according to available publishers in the system.