On-Demand Delivery of Large Popular (Multimedia) Data Files

点击翻译按钮获取中文摘要

This proposal will develop next generation data delivery techniques for the Web and other on-demand data delivery systems that contain large, popular, widely shared data files. The project includes several research efforts. One key goal is to design improved (or optimized) delivery architectures that provide immediate service at reduced (or minimal) server bandwidth. Another key goal is to develop tractable models to compute the regional cache contents that minimize delivery cost for systems with diverse object lengths and delivery rates, and with heterogeneous regional server capabilities and client loads. A third goal is to derive basic system design principles by using the optimized cache content calculations to determine how delivery cost and the optimal caching strategy are influenced by key system and workload parameters. A fourth goal is to develop extensions to the dynamically scheduled delivery architecture to optimize support of (encoded) variable bit rate data streams. Finally the project will address several implementation issues, including server disk layout for the dynamically scheduled data segments, methods for informing clients of scheduled and reassigned multicast segment deliveries, and methods for coordinating shared data delivery between the (remote) information source server and the regional/proxy server. The results of this research are being incorporated in a prototype on-demand data delivery system to demonstrate implementation feasibility and to validate the projected performance improvements.