自相似网络流量的处理和分析  

Analysis and processing of self-similar network traffic data

作　　者：吕军[1] 

机构地区：[1]清华大学信息网络工程研究中心,北京100084

出　　处：《清华大学学报（自然科学版）》2008年第7期1186-1189,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家"八六三"高技术项目(2005AA112130)

摘　　要：为了解决自相似模型难以进行自相似网络流量趋势预测的问题,提出时间序列分析中短时相关模型(自适应自回归模型)的方法用于流量数据的估计;同时为了提高预测精度,提出改进的最小平方格型(modified least squarelattice,MLSL)算法,使模型参数不断递推修正,收敛到最佳值。仿真试验结果验证了短时相关模型在网络流量预测应用中的可行性,实现了自相似网络流量的短期预测,该算法比最小平方(least square,LS)算法均方误差减少20%,具有收敛快、预测精度高的优点,而该算法的计算量减少一半。An adaptive auto-regressive model based on time serial analysis theory was developed for network management, such as forecasting of network traffic with self-similar characteristic, which is difficult to analyze with self-similar model. A modified least squares lattice （MLSL） algorithm was used to improve the prediction accuracy. In this algorithm, the model parameters are recursively updated with each new input data and can converge to the optimal values. Simulations validate the feasibility of the short term memory model for network traffic forecasts and show that network traffic can be accurately forecast in short time frames. The mean error of the MLSL algorithm is 20% better than that of the least squares algorithm with half the computational cost.

关 键 词：计算机网络 自相似特性 流量模型 短时相关模型 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]