The Research On Variable-bit-rate Video Traffic Prediction And Dynamic Bandwidth Allocation

With the popularization of broadband network, multimedia applications are developing rapidly. Thus, network bandwidth allocation and management become more and more complex and challengeable than ever before. Variable-bit-rate (VBR) video traffic is one of the primary and most complex traffic generated by multimedia application. Network control, on one hand, must support different types of traffic classes to meet QoS requirements. On the other hand, it is needed to utilize the network resources efficiently. This paper focuses on VBR video traffic prediction and dynamic bandwidth allocation based on the predictive results.Firstly, the characteristics of VBR video traffic and the status of researching on traffic prediction are introduced. Three kinds of prediction models which are short-memory models, long-memory models and neural network models, are discussed. Based on above, fuzzy neural network and recurrent neural network are used to predict VBR video traffic. The output of predictors are combined by using three different combination schemes which are arithmetic averaging, recursive least square algorithm and MLP neural network, to enhance prediction accuracy. The different combination schemes are compared and evaluated. The experiment results show that the expected performance is achieved. Secondly, the prediction for VBR video traffic based on FARIMA model is discussed in the paper. And two different schemes, least mean square error (LMS) algorithm and least mean kurtosis (LMK) algorithm are analyzed. Traffic Prediction experiment on different time scales shows that FARIMA outperforms the traditional models, especially for certain time scale is very satisfied. Furthermore, the experiment shows that the prediction accuracy and performance of LMK can be improved significantly at a small extra computation cost compared to LMS.Finally, the dynamic bandwidth allocation scheme for the VBR video traffic...