Network traffic prediction is very important in network protocol designing

network management and high performance routers designing etc.Currently

ARMA and FARIMA time series are the main models used to fit and predict the network traffic.But the relation between time scale and time series models hasnt been studied.The network traffic was modeled according to different time scales using the traffic trace data taken from the Internet traffic archive

and the prediction performance of those models was compared.The results show that time series models perform better under the large scale(minute)than under small time scales(millisecond and second)

and the performance of self similarity model FARIMA shows no advantage over other models.