Summary: Resource Prediction for Admission Control of Interactive Multimedia
Silvia Hollfelder and Karl Aberer
GMD - IPSI, Dolivostr. 15, 64293 Darmstadt, Germany
Interactive multimedia sessions have uncertain, varying consumption rates due to users' interactive behavior. In this paper, we
propose two approaches for prediction of consumption rates required for admission control. They differ in the assumptions of
the available knowledge: The observation-based approach predicts the future consumption from the past system behavior. The
stochastical approach requires a user behavior model from which the prediction is deduced. We discuss the pros and cons of
the two approaches with respect to available information sources, accuracy of prediction, and suitable scenarios.
Keywords: interactive multimedia applications, user modeling, predictive service
In highly interactive applications a user controls the course of a running session in terms of which media(-
combinations) will be presented at what time. Such applications play a key role in domains like home entertainment
(e.g., news-on-demand, interactive VoD, action games), home shopping (e.g., product catalogs), education and
training (e.g., interactive computer-based training - CBT), tourist information (e.g., virtual travel guides), and
multimedia production (e.g., video editing).
Multimedia systems have to provide mechanisms that are able to deal with the special characteristics of such
applications to enforce QoS parameters like continuous playout of media data and low start-up latency between