A major problem with the Internet and web-based applications is the scalable delivery of data. Lack of scalability can hinder performance and decrease the ability of a system to perform as originally designed. One of the most promising solutions to this scalability problem is to use a multiple channel hybrid data dissemination server to deliver requested information to users. This solution provides the high scalability found in multicast, with the low latency found in unicast. A multiple channel hybrid server works by using a push-based multicast channel to deliver the most popular data to users, and reserves the pull-based unicast channel for user requests and delivery of less popular data.
The adoption of a multiple channel hybrid data dissemination server, however, introduces a variety of data management problems. In this dissertation, we propose an improved multiple channel hybrid data dissemination model, and propose solutions to three fundamental data management problems that arise in any multiple channel hybrid scheme. In particular, we address the push popularity problem, the document classification problem, and the bandwidth division problem. We also propose a multicast pull channel to the common two-channel hybrid scheme. Our hypothesis that this new channel both improves scalability, and decreases variances in response times, is confirmed by our extensive experimental results. We develop a fully functioning architecture for our three-channel hybrid scheme. In a real world environment, our middleware is shown to provide high scalability for overloaded web servers, while keeping the response times experienced by clients at a minimum. Further, we demonstrate that the practical impact of this work extends to other broadcast-based environments, such as a wireless network.
