Service Oriented Bandwidth Estimation And Bandwidth Allocation

With the increasing of subscribers and arising of VBR service, the network flow with bursty characteristic increases rapidly. In order to guarantee that network can provide stable and reliable service for users while satisfy QoS of different services, many researchers have studied flow control strategies of various networks. As the basis study of network flow control strategy, flow prediction and bandwidth allocation methods are important subjects of network flow engineering.This paper proposed network flow prediction and bandwidth prediction methods where network flow prediction method predicts network flow based on time series model which can describe dynamic of series and development trend while bandwidth prediction method predicts bandwidth of service with statistic characteristics and QoS demands of service.The proposed network flow prediction method is based on FARIMA model. Since both the actual network flow and the flow generated by FARIMA model have the property of self-similar, we first estimate the parameters of FARIMA model according to the actual network flow and then predict network flow with the generated series of the fitted FARIMA model. This method utilizes d order difference to convert parameter estimation of FARIMA model to parameter fitting of ARMA model. We introduce model parameter judgment mechanism in multi-step forecast of network flow to decrease parameter fitting steps, which reduces the complexity of the network flow prediction method. After comparing the forecast value and the actual flow, we can see that this method can predict the change of network flow effectively.The proposed bandwidth prediction and allocation method present the mapping relationship between delay tolerance and buffers length of queueing system on the basis of buffer overflow probability formula derived by Norros. The method analyze how mean value, variance coefficient, Hurst parameter, delay and loss probability will affect the prediction value. Given for the bursty of traffic flow, the predicted bandwidth can meet QoS requirement of services.In addition, we analyze the self-similar property of network flow and give common methods to estimate Hurst parameter. We estimate the Hurst parameter of actual network flow collected at GGSN port and verify the self-similar property of network flow. On the basis of the study of self-similar traffic flow model, we generate the self-similar series of FGN and FARIMA model with RMD and Hosking methods and derive the self-similar series of ON/OFF model where lasting time of each state is Pareto distribution. Through estimating the Hurst parameter, we verify the self-similar property of above generated series.In conclusion, we propose network flow prediction and bandwidth prediction methods which can be used to forecast the change of network flow and service bandwidth based on time series model, flow characteristics and QoS demands of services. Meanwhile, this paper does some foundational studies about Hurst parameter estimation and self-similar series generation.