摘要
自相似性是网络的普遍属性,并且对网络性能具有重要影响。在网络性能研究中,利用自相似的长程相关性来进行预测对于有效分配和利用网络资源以保证网络QoS及提高网络性能是非常有意义的。然而,由于自相似通信量同时具有长程相关和短程相关性的多尺度性和非线性使得通信量的预测非常困难。文章在充分考虑自相似网络通信量这些特性的基础上,提出了利用人工神经网络来进行预测的方法。我们首先根据研究目标构建了多时间尺度预测的人工神经网络,并且对输入/输出向量处理、参数选择和学习算法进行了讨论;然后,我们利用FARIMA为模型合成的同时具有LRD和SRD性质的通信量trace进行了多尺度预测的实验研究,结果表明可以利用该算法进行多时间尺度预测,这对于优化网络控制策略是非常有意义的。
Serf-similarity is a ubiquitous phenomena spanning across diverse network environments and has great effect on network performance.The Long-Range Dependence（LRD） structure in self-similar network traffic could be exploited in traffic prediction,and it is very useful in resource allocation.But the traffic prediction is very difficulty because of its multi-scale and non-linear feature.Having considered these features of self-similar network traffic,the prediction algo- rithm with ANN is proposed in this paper.At first,ANN for the multi-scale traffic prediction is constructed.Its I/O vectors procession,parameter selection and training scheme are also discussed.Then the artificial traces are generated with Fractional-ARIMA model and are used in the experiments of ANN multi-scale prediction.The result shows that this algorithm can predict the self-similar network traffic at multi-scale.It is very useful to optimize network control scheme and enhance network performance.
出处
《计算机工程与应用》
CSCD
北大核心
2005年第28期26-28,共3页
Computer Engineering and Applications
基金
国家自然科学基金重点项目子项目资助(编号:60132030)
关键词
自相似
通信量
多尺度
预测
self-similar,network traffic,multi-scale,prediction
