Life-Cycle Considerations for Wide-Area Distribution of Multimedia Data

Abstract

Various applications for world-wide multimedia object distribution arise currently as more and more companies integrate streamed continuous media elements in their web pages and as TV broadcasters add computer programs and data to their video streams.Currently, the streams that can be found on the web have only very small data volume, but as network capacities increase, the data volume mat is used by applications will also be increased. TV applications have a lack of networking support and are unable to give feedback to the sender. In the future the integration of computer networks, telephony and broadcast networks will achieve a functional integration of these two worlds.

But as the amount of continuous media data which is offered for everybody to retrieve increases, simple caching mechanisms that work currently well will not be appropriate any more. To address this danger, the introduction of distribution mechanisms which provide more coordination among caches is necessary. As a basis for such a development, we investigate the probability of movies to be requested. This is chosen as a starting point because the current simple caching mechanisms operate very successfully for small-volume data such as text or icons, and high-quality movies are currently the most attractive medium which has the very high requirements in terms of storage as well as bandwidth. In this paper, we review current models on access probabilities and we give strong indication that the current assumptions about hit rates are invalid if the number of users connecting to an individual server differ widely. Since on-demand retrieval of large continuous media streams (especially movies) is currently available only as music CD and movie rental or sale, we take that data as a basis to create a model for access probabilities on this most storage-and bandwidth-demanding multimedia data.