计算机工程

• 移动互联与通信技术 • 上一篇    下一篇

混合SDN的自适应流量估计方法

李晓方1,2,王子磊1,奚宏生1   

  1. (1.中国科学技术大学未来网络实验室,合肥 230027; 2.中国酒泉卫星发射中心,甘肃 酒泉 732750)
  • 收稿日期:2015-03-19 出版日期:2016-03-15 发布日期:2016-03-15
  • 作者简介:李晓方(1989-),男,硕士研究生,主研方向为软件定义网络、信息安全;王子磊,副教授;奚宏生,教授。
  • 基金项目:

    国家自然科学基金资助项目(61233003,61203256);国家“863”计划基金资助项目(2014AA06A503)。

Adaptive Traffic Estimation Method for Hybrid SDN

LI Xiaofang  1,2,WANG Zilei  1,XI Hongsheng  1   

  1. (1.Laboratory of Future Network,University of Science and Technology of China,Hefei 230027,China; 2.Jiuquan Satellite Launch Center,Jiuquan,Gansu 732750,China)
  • Received:2015-03-19 Online:2016-03-15 Published:2016-03-15

摘要:

流量矩阵是混合软件定义网络(SDN)流量工程的重要输入,但难以全部直接测量,已有的估计方法主要针对传统IP网络,不完全适用于混合SDN网络。针对该问题,提出一种基于源-目的(OD)流聚类的自适应多Elman神经网络算法。通过对OD流按照时间变化模式进行聚类,将单一的高维训练样本分解为多个低维训练样本,强化各低维样本的关键特征,以训练相应的Elman神经网络,构成多Elman神经网络模型,并利用混合SDN中部分OD流可以持续精确测量的特点,根据网络状态变化动态调整估计算法的参数。实验结果表明,与广义层析重力算法相比,该算法具有更高的估计精度和更好的自适应能力。

关键词: 软件定义网络, 流量矩阵估计, 神经网络, 聚类, K均值, 自适应性

Abstract:

Traffic matrix is an important input for traffic engineering in hybrid Software Defined Network(SDN),but it is difficult to be directly measured.Existing estimation methods are mainly for the traditional IP networks,not fully applicable to hybrid SDN.In order to solve this problem,this paper proposes an Adaptive multi-Elman Neural Network Algorithm(AMElman) based on Origin-Destination(OD) flow clustering.By clustering OD flows according to the time change pattern,the single high-dimensional training sample can be decomposed into several multiple low-dimensional training samples,and each of the low-dimensional samples’ key characteristics can be strengthened to train the corresponding Elman neural network and construct the multi-Elman model.At the same time,using the feature that part of the hybrid SDN OD flows can be measured accurately and constantly,the estimation algorithm’s parameter can be dynamically adjusted based on change of network status in the estimation process.Experimental results show that this algorithm has higher estimation accuracy and better adaptability than General Tomogravity(GT) algorithm.

Key words: Software Defined Network(SDN), Traffic Matrix(TM) estimation, neural network, clustering, K-means, adaptivity

中图分类号: