Algorithms for Media-on-Demand Systems

本研究涉及媒体点播系统。 媒体点播系统是一种分布式网络系统，其中服务器响应客户端的请求以接收各种类型的媒体，诸如视频、音频或大文件。 这样的系统可以在像酒店或大学这样的组织内实现，或者通过互联网实现。 目前，互联网媒体点播系统使用单播，这意味着每个客户端接收其自己的传输。 这种方法不能很好地扩展流行媒体，因为在流行媒体中，可能会在短时间内收到数十万个请求。 因此，今天，例如，提供电影点播的酒店通常会让客户等待长达一个小时的开始时间。 在这项研究中，主要研究者研究算法的系统中，多播或广播是可用的和缓冲存储器是足够的。 研究了服务器带宽、启动延迟、客户端接收带宽和客户端缓冲区大小之间的权衡。 特别是，新的算法和数学分析开发，以探索这些权衡。 研究的主要目标是了解如何用尽可能少的资源实现最高的服务质量。

This research involves media-on-demand systems. A media-on-demand system is a distributed network system where servers respond to demands by clients to receive various types of media such as video, audio, or large files. Such systems may be implemented within an organization like a hotel or a university, or over the Internet. Currently, Internet media-on-demand systems use unicast, meaning that each client receives its own transmission. This method does not scale up well for popular media where hundreds of thousands of requests might be received in a short time. Thus today, for example, hotels providing movies-on-demand typically make customers wait up to an hour for starting times. In this research, the principal investigator studies algorithms for systems where multicast or broadcast is available and buffer storage is adequate. Tradeoffs among server bandwidth, startup delay, client receive bandwidth and client buffer size are studied. In particular, new algorithms and mathematical analyses are developed to explore these tradeoffs. The main goal of the research is to understand how to achieve the highest Quality of Service with as few resources as possible.