Automatic prediction for IP backbone network traffic

Abstract

Efficient and reliable network traffic prediction is the basis of network planning and capacity expansion construction.Currently,there is no integral theoretical model to describe internet traffic.Most of the industry designs simplified and operable prediction models.Firstly,according to the characteristics of China Telecom’s IP backbone network traffic and its planning requirements,the IP backbone network traffic was analyzed and forecasted by using the multi-factor regression model and the function adaptive mode of time series.The characteristics,advantages,disadvantages and applicable scenarios of these two models were compared based on simulation of a large number of actual network data.A set of principles and methods for selecting prediction model and optimizing parameters were proposed.Then,an automatic forecasting system with the high performance of dealing with hundreds of time series was built to greatly simplify and improve the traffic prediction efficiency.Finally,the development orientation of network capacity extension and key points of future IP traffic prediction were prospected.

Keywords

• time series
• traffic forecast
• prediction model