自相关性对IP网络数据流量预测影响的实验分析

为了测定自相关性对于IP网络数据流量预测的影响,利用局域网和广域网上的真实流量数据集,设计并进行了流量聚合实验和洗牌实验.得到的实验数据显示,随着聚合尺度的增大,平均相对预测误差减小;自相关性越强,预测误差越小;上述影响主要来自于短时相关性.实验结果表明,自相关性对于数据流量预测具有重要影响,自相关性越强,预测效果越好;而且,短时相关性的影响是主要的,长时相关性的影响很小.

Experimental analysis of auto-correlation's effects on traffic prediction of IP network data

In order to measure the effects of auto-correlation on traffic prediction of IP network data, traffic aggregation and shuffle experiments were designed and conducted with the real traffic traces of LAN and WAN. The following results were obtained from the experimental data. The average relative prediction error descended with the increases of the aggregation level. The stronger the auto-correlation, the smaller the prediction error. The above effects were mainly from the short-range dependence. The results show that auto-correlation has important effects on data traffic prediction. The stronger the auto-correlation, the better the prediction performance. Furthermore, the effects from short-range dependence are dominant and the effects from long-range dependence are very little.