An integrated approach to end-to-end analysis of distributed video-on-demand systems
P. Mundur, Arun K. Sood, Robert Paul Simon · 2000
Distributed Video-on-Demand (VoD) systems are expected to be one of the most important services supported by the next generation of high-speed networks, video servers, and distributed multimedia file systems. Using a distributed VoD system, a client can request a video from anywhere and at any time. In response to a client's request, VoD systems deliver high quality digitized video directly to client set-top boxes. A typical VoD architecture consists of three subsystems: storage, network, and client. Admission control tests are employed on all subsystems before admitting a new request to check whether resources are available to provide guaranteed service. This architectural approach represents a dramatic break from a best-effort environment. From a research point of view, VoD systems pose great challenges in the areas of retrieval and delivery for real-time performance. Much of previous research restricts the analysis of VoD systems to either the storage or the network subsystem. The focus of my dissertation research is the end-to-end analysis of a distributed VoD system. I analyze the distributed architecture of the VoD system to design global request handling and admission control strategies and evaluate them using global metrics. I also analyze and model all subsystems—server, network, and client—involved in providing guaranteed services to an individual request in an integrated framework. The objective of this analysis is to develop a methodology for determining efficient ways of using all resources in the VoD architecture within the constraints of providing guaranteed high quality services to each request. In this dissertation, I develop an analytical model that integrates the video server retrieval techniques with the network transfer mechanisms by relating buffer management at the server with rate-based scheduling at the network. The significance of this integrated analysis is the development of a method for designing and measuring the effectiveness of end-to-end admission control and request handling policies. I extend the analysis to include the effects of controlling the network delay jitter, where jitter is the difference between maximum and minimum end-to-end delay experienced by packet transfers. I prove that the client buffer requirements will be reduced if the underlying network service discipline provides a non-trivial delay jitter bound. I propose a new technique called threshold-based admission control for VoD systems and develop an analytical model for computing blocking probabilities efficiently. Using simulation, I show that threshold-based VoD system behavior conforms to the analytical model. Through analytical and simulation results, I show that threshold-based admission control can be used to manipulate VoD performance by restricting access to lower priority request classes. I analyze the integrated admission control strategy through an extensive simulation of a distributed VoD system consisting of server clusters and networks. My results show that request handling policies based on limited redirection of blocked requests to other resources perform better than load sharing policies. This shows that limited redirection must be incorporated in the reservation or signaling protocols. For providing service at different quality of service levels, I show that the downgrade option coupled with threshold restrictions is a powerful tool for transforming an incoming request mix into a workload that the VoD system can handle.