小时间粒度网络流量自回归预测分析

网络流量测量和预测是网络QoS管理和流量工程中一个重要的组成部分,尤其是对于为了保证网络QoS而引入的一些实时方法,比如接纳控制,资源预留等。较好的网络流量预测效果,能有效地提高这些方法的工作效率,从而有效提高网络带宽的利用率,保证网络QoS。所以高效的网络流量预测不仅是值得的,而且是必要的。由于许多文献研究的是网络流量在大时间粒度(天、周、月等)上的自回归特性,不能用于这种以秒级为单位的接纳控制、资源预留等实时方法,所以本文具体分析了网络流量在小时间粒度的自相似特性,并提出了其自回归预测模型。在模拟实验中采用了实际网络流量,并证明了在大多数情况下预测误差小于15%的概率为90%,它可有效地应用到接纳控制等方面的网络流量预测中。

An Analysis of Traffic Load's Auto Regressive Prediction in Small Time Granularity

Traffic load measurement and prediction is an important component of network Quality of Service(QoS)management and traffic engineering.Especially to some real time methods in order to ensure QoS,such as Admission Control and Resource Reservation and so on,better traffic load prediction result can improve their work efficiency greatly and deeply advance network bandwidth utilization and ensure better QoS.So we regard that efficient and effective traffic load prediction techniques are desirable necessary.Much former research work is analyzing traffic load auto regressive characteristic in large time granularity,such as day,week or month and so on,but they couldn't be used in these real time methods including admission control and resource reservation.So we analyze the self-similarity of traffic load in small time granularity and propose an Auto Regressive prediction model.In the simulation,we adopt the real traffic load of NLANR and have proved that the probability of prediction error less than 15% is about 90%.