Analysis of End - to - End Admission Control

iv 1 Introduction 1 Evolution of standards and protocols for high speed networks coupled with rapid advances in the storage technology has spurred interest in distributed multimedia applications. Distributed video-on-demand systems are expected to be one of the most important services supported by the next generation of high-speed networks and video servers. Because of the stringent real-time performance and synchronization requirements of video applications, there has to be a tight coordination among all components of a VoD system. Current data management practices both at the server and network levels have to be modiied to address real-time requirements of multimedia data. We tackle the issue of tight coordination of all system components as a resource allocation problem and develop admission control policies to optimally allocate resources from le servers to networks to end-user presentation devices. The primary goal of our research is to develop optimal and eecient end-to-end admission control policies for video-on-demand (VoD) systems. We adopt a uniied model approach which considers all subsystems, storage and network in solving the end-to-end problem. In contrast, most of previous research has focused on analyzing isolated subsystems, mostly the storage system. Using an analytical framework, we have developed a recursive algorithm that ee-ciently computes blocking probabilities at the video server. We have used the analytical model to evaluate threshold-based admission control policies for a multi-class VoD system. Threshold-based analysis provides a convenient way of introducing the concept of priorities among request classes in our queueing network model which otherwise would have proven to be computationally intractable. We have shown through numerical analysis the eeect of diierent thresholds on overall system blocking rates. We propose to extend the analytical model to include all aspects of a distributed VoD architecture, multiple classes of requests, network segments, many video servers and archive servers. We will investigate into nding an optimal mix of admission control policies running on each shared resource that would form the basis for optimizing the overall VoD architecture. We will investigate if it is feasible to extend quantitative techniques to any arbitrary network. We propose to conduct a simulation to validate our analytical results.