Gmd Report 40 Resource Prediction and Admission Control for Interactive Video Browsing Scenarios Using Application Semantics

Highly interactive multimedia applications, like browsing in video databases, generate strongly varying loads on the media server during the presentation of media data. Existing admission control approaches for limiting the number of concurrent users and thus guaranteeing acceptable service quality are only suited for applications with uniform load characteristics like video-ondemand. We propose an approach to admission control that is based on the stochastical model of Continuous Time Markov Chains, which allows to describe the different presentation states occurring in the interactive access to the multimedia database. The model is derived from semantic information on the forthcoming browsing session, in particular, it considers the relevance of the videos to the user. In this way a more precise prediction on resource usage can be given for achieving the two goals of Quality of Service (QoS) and good server utilization. The admission control mechanism is part of a multimedia database architecture for supporting efficient browsing in large video collections.