Performance Evaluation of Video-on-demand Service for E-learning in a Campus Network

Since it is inconvenient to generate a large number of video requests or changing network factors to perform actual network measurements, we have instead developed an analytical model for evaluating the network queueing delay. Our modeling approach is based on the use of Maximum Asymptotic Variance (MVA) technique to predict the maximum number of video flows that can be supported at a single node for a given delay bound requirement. Then, we extend the prediction from the single queue case to the multiple queue case. Network measurement is then performed to validate the model accuracy. With the analytical model, we predict the number of admissible flows and investigate the effects of delay bound requirements and the network paths on the network performance. This paper evaluates the possibility of VoD service deployment in the KMUTT campus network. Particularly, we are interested in the number of video flows that can be supported along the paths from the video server to end-user workstations located around the campus. Our evaluation approach is based on the Maximum Asymptotic Variance (MVA) technique, which predicts the number of admissible flows at a single node for a given delay bound requirement. We then apply a simple probabilistic bound to deal with multiple nodes. Network measurement using a Distributed Benchmark System (DBS) is performed to validate the model accuracy. In our experiments, MPEGcoded video traffic recorded from class lectures is used. Effects of delay bound requirements and the network paths are investigated. Based on the results, we found that the link capacity may have to be upgraded or some mechanism to prioritize the video traffic is needed. The paper is organized as follows. In Section II, related research on analytical modeling of queueing performance techniques as well as the MVA approach are reviewed. In Section III, we describe the topology of our campus network, the network factors, and characteristics of the MPEG traces under study. In Section IV, we present a simple extension to the MVA approach to deal with the multiple queue case. The model accuracy is validated by comparing to the measurement results obtained from Distributed Benchmark System (DBS). In Section V, the results from the analytical model are then reported. The conclusion is given in Section VI.