利用数据挖掘技术改进TCP CUBIC拥塞控制算法  

Improved algorithm for TCP CUBIC congestion control based on data mining technology

作　　者：张錋 毛澍 李彦庆 张晶晶 武宏斌 韩啸 Zhang Peng;Mao Shu;Li Yanqing;Zhang Jingjing;Wu Hongbin;Han Xiao(Institute of Information&Communication,Global Energy Interconnection Research Institute,Beijing 102209,China;State Grid Key Laboratory of Information&Network Security,Global Energy Interconnection Research Institute,Beijing 102209,China)

机构地区：[1]全球能源互联网研究院,信息通信研究所,北京102209 [2]全球能源互联网研究院,信息网络安全国网重点实验室,北京102209

出　　处：《计算机应用研究》2018年第10期3044-3047,共4页Application Research of Computers

摘　　要：TCP CUBIC拥塞控制算法无法主动判断拥塞和预测丢包。为解决这一难题,算法首先采集网络节点数据,利用多维关联挖掘方法建立模型,引用RTT最大值和最小值变化率判定网络拥塞;然后划分拥塞等级,利用动态神经网络算法训练模型,确定拥塞窗口的立方增长因子λ和线性增长参数S数值大小,制定不同策略主动调整拥塞窗口增长速度,利用贝叶斯学习推理策略,动态学习预测连接路径丢包概率;最后,结果表明改进算法能有效调整拥塞和预测丢包。TCP CUBIC congestion control algorithm cannot take the initiative to judge congestion and predict packet loss.In order to solve this problem,firstly the algorithm collected the network node data,established the model by using the multi-dimensional association mining method,and used the RTT maximum value and the RTT minimum value change rate to determine the network congestion.Secondly,it divided the congestion level,used the dynamic neural network algorithm to train the model,determined the cube growth factorλand the linear growth parameter S value of the congestion window,and developed different strategies to adjust the growth speed of the congestion window.It used the Bayesian to learn reasoning strategy,predicted the probability of connection path loss dynamically.Finally,the results show that the improved algorithm can effectively adjust the congestion and predict packet loss.

关 键 词：数据挖掘 关联挖掘 拥塞控制 往返时延 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]