项目名称: 软件定义网络(SDN)环境下基于机器学习的路由预规划研究
项目编号: No.61502106
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 自动化技术、计算机技术
项目作者: 郑相涵
作者单位: 福州大学
项目金额: 21万元
中文摘要: 在软件定义网络(SDN)环境下,控制器需要为新一个新流制定并安装路由策略,这给控制平面与数据平面造成巨大的计算与通信负担,并不高效。在此背景下,智能化的路由预规划被认为是路由有效性提升的重要途径。本项目面向SDN路由预规划领域,立足于机器学习理论,从流特征提取、需求预测、路径规划三方面依次开展研究。首先,根据用户历史数据,分析数据包与数据关联特征,引入半监督式聚类算法,研究SDN流的高效归类、粒度分析与特征提取,为路由预规划提供前提;在此基础上,引入监督式分类算法,结合用户与数据平面负载特征,开展SDN流的需求预测研究,为路由预规划的策略制定提供支撑;最后,制定相应的流业务处理、数据平面拓扑裁剪、多约束因子权重分配等策略,并据此设计面向需求的个性化路径规划方法,实现最优化的路径选择。研究成果将结合理论与应用实际验证,确保可用性、可靠性与有效性。
中文关键词: 软件定义网络;路由预规划;机器学习;特征提取;特征学习
英文摘要: In Software Defined Network (SDN) environment, controller has to compute and install routing strategy for each new flow, leading to a lot of computation and communication burden in both controller and data planes. In this background, intelligent routing pre-design mechanism is regarded to be an important approach for routing efficiency enhancement. This project investigates and proposes efficient SDN routing pre-design solution from three aspects: flow feature extraction, requirement prediction and route selection. Firstly, we analyze and extract data packet and association features from user history data, apply these features into semi-supervised clustering algorithm for efficient data classification, analysis and feature extraction; After that, flow service requirement could be predicted through extraction of user, flow and data plane load features and implementation of supervised classification algorithm; furthermore, we propose corresponding handling strategies related to data plane topology, flow forwarding and multi-constraint weight assignment, and proposes personalized routing selection mechanism. The research output will be implemented and evaluated in testing bed for insurance of availability, reliability and efficiency.
英文关键词: Software Defined Network;Routing Pre-Design;Machine Learning;Feature Extration;Feature Learning