Content area
Abstract
Due to high bandwidth requirement and rate variability, delivering video across wide area networks (WANs) is a challenging issue. By caching video data, a video proxy server (proxy) close to the clients can assist the delivery. In our setting, a video server is connected to a proxy via WANs and a proxy reaches clients through Local Area Networks. The proxy allows partial video caching and stores a certain number of video frames. Since there are two video sources, video data has to be synchronized before playback at a client. Two delivery models, client-synchronization and proxysynchronization, in terms of where to synchronize data are identified. They have different complexity and resource consumption in the proxy and the client.
The thesis first focuses on the fundamental understanding on delivering a single video using the two models and investigates the effectiveness in WAN bandwidth reduction. We study the tradeoffs between client buffer, start-up delay, proxy resource consumption, and WAN bandwidth requirement. Given a transmission rate for WANs, we propose several frame selection algorithms to determine cached video frames. In client-synchronization model, a scheme, which partitions a video into segments of frames, is shown to offer the best tradeoff. In proxy-synchronization model, caching the initial portion (prefix) of a video is adopted. This approach is proved to provide minimum space requirement in the proxy storage.
Then dynamic adjustment is considered for multiple videos. The purpose is to design a caching policy suitable to a video proxy server and thus multiple-level video caching is introduced. For each video, the proxy caches a smaller amount of data at a lower level or accumulates more to reach a higher level. The proxy dynamically adjusts the cached data by choosing an appropriate level based on network condition or the popularity of the video. A frame selection scheme is proposed to decide cached frames for different levels. A multiple-level video caching policy is also designed to efficiently utilize the proxy resources to maximize the overall WAN bandwidth reduction. We evaluate the policy by investigating the situations in which the user access pattern is priorly known or unknown.