In recent years there has been an explosion of research in packetized VBR video, with the great majority of the work addressing live video such as video conference and the broadcast of a sporting event. While this research on live video certainly merits the attention it has received, we feel that the majority of the video carried on high-speed networks will emanate from prerecorded sources. These sources include full-length movies, music video clips, and educational material. From the perspective of the transport network, prerecorded VBR video sources are fundamentally different from live video sources: for live video, the exact dynamics of the VBR traffic are unknown; for prerecorded sources, the traffic is fully known before it is transmitted into the network. Surprisingly, little attention to date has been given to the management of prerecorded video traffic in high-speed networks. There is an emerging type of prerecorded traffic that will likely dominate the broadband residential networks of the future --- Video-on-Demand (VoD) traffic. It is expected that the telecommunication networks of the future will offer a rich variety of VoD services. For example, a costly service would allow the movie to instantaneously begin at the viewers request and would provide full VCR control (pause, rewind, and fast-forward). An inexpensive service would require the viewer to schedule the commencement of the movie minutes or hours in advance and provide no VCR control. This research will develop, analyze and compare a wide variety of end-to-end traffic management schemes for prerecorded video. A central theme of the research is that VoD traffic is prerecorded, and that video providers, networking companies, and equipment manufacturers should fully exploit this important property. Traffic management schemes will be designed which allow varying degrees of VCR control, schemes which guarantee zero cell loss, and schemes which permit infrequent cell loss. Initially, the research wi ll assume that an ATM network transports the video directly between the provider's server and viewer's set-top box. But prerecorded traffic management schemes will also be developed for architectures with neighborhood caches and for architectures based on future generations of the Internet. This research will also investigate how the traffic management schemes impact the resource management of the provider's VoD server, and how VoD services should be priced. The research will lead to a comprehensive methodology for the analysis of prerecorded video over packet-switched networks. This research is described in greater detail on the following Web page: www.seas.upenn.edu/~ross/nsf/prerecorded.html .
