Stochastic Resource Prediction and Admission for Interactive Sessions on Multimedia Servers

摘要

In highly interactive multimedia applications startup latency is significant, and may negatively impact performance and Quality of Service (QoS). To avoid this, our approach is to admit whole multimedia sessios instead of single media streams. For the prediction of the varying resource demands within a session, which are mainly correlated to user behavior, we model user behavior as Continuous Time Markov Chains (CTMCs) In this paper, we preopose a mathematical analysis of the CTMC model. This allows to anticipate possible overload and in turn to plan an admission control policy. As a result, our approach provides better control on the tradeoff between server utilization and QoS. Simulation studies confirm this capability.