Abstract: In BoD(Bandwidth on Demand) protocol,full utilization of remaining bandwidth after allocation on demand can reduce service delay．First on account of how to make efficient use of remaining bandwidth and its effect on feedback control,this paper proposed a new BoD bandwidth request allocation algorithm based on prediction and on-board control．Secondly in view of both the long and short range dependence of network flow,we proposed ARFIMA(p,d,q) model for traffic modeling and prediction,which can represent both long and short range dependence according to the different values of parameter d．Finally we employed OPNET for simulation system．The simulation results indicate that,for both long and short range dependence service flow,BoD request allocation scheme based on ARFIMA(p,d,q) prediction and on-board feedback control can reduce service delay to the most degree,with the help of remaining bandwidth.

Key words: BoD,ARFIMA(p,d,q) model,Traffic prediction,Self-similar,Long-range,OBP