University of Toronto, 2010.
Mission-critical distributed applications, such as Internet advertising platforms, increasingly utilize distributed publish/subscribe systems as a messaging substrate for information dissemination. These applications require low latency performance from the substrate, as the timely delivery of messages can have a direct on impact revenue. The cost of managing and operating publish/subscribe systems, however, can be prohibitive due to system size and scale. It is, therefore, critical to derive low latency message delivery from a minimal set of system resources.
To this end, this thesis presents a solution for designing low latency, minimal-broker overlay networks for content-based publish/subscribe systems. The solution includes a framework for quantifying the similarity of clients and brokers, and algorithms for constructing overlay topologies where brokers sharing similar interests are assigned a direct overlay connection. Additionally, a load model and algorithms are presented for designing overlays that utilize a minimal number of brokers in order to reduce system cost.