基于OCTEON多核处理器的网络流量分类技术研究与实现

发布时间：2018-09-19 08:18

【摘要】：随着互联网技术的高速发展,对于网络环境安全的需要也越来越紧迫。网络流量分类技术作为实现网络可控性的基础技术和重要手段,在网络管理、服务质量保障和网络安全等领域正在发挥越来越重要的作用。本文研究了现有的网络流量分类技术,总结了基于端口号匹配、基于特征字段分析、基于传输层行为模式和基于流统计特征的机器学习方法各自的优势和不足。鉴于基于机器学习算法的网络流量分类技术分类精度高、易于扩展等优点,选取随机森林作为网络流量分类方法,并针对传统随机森林的不足提出了改进策略,最后再将改进的随机森林策略实现到OCTEON平台上,设计并实现了基于OCTEON多核处理器的网络流量分类系统。最后,搭建了系统测试环境,对系统进行了有效性测试。主要工作有以下几个方面:1、在充分理解传统随机森林算法的基础上,根据算法本身存在的单节点模型训练速度慢和不区分基分类器强弱性能差异的不足,从并行性和加权调整两方面入手提出了改进的随机森林策略;2、对OCTEON本身的硬件特性进行研究,将OCTEON的运行模式、核间同步通信机制和内存分配机制等优秀特性与改机的随机森林策略结合起来,在OCTEON平台上实现改进的随机森林模型;3、实现了基于OCTEON多核处理器的网络流量分类系统,详细介绍了系统的架构设计,各个功能模块的详细设计和实现,并搭建了系统测试环境,对系统的分类性能进行了测试,验证了系统在流量分类方面的优异性能。
[Abstract]:With the rapid development of Internet technology, the need for network environment security is becoming more and more urgent. As the basic technology and important means to realize network controllability, network traffic classification technology is playing a more and more important role in network management, quality of service assurance and network security. In this paper, the existing network traffic classification techniques are studied, and the advantages and disadvantages of the methods based on port number matching, feature field analysis, transport layer behavior pattern and flow statistics feature based machine learning are summarized. In view of the advantages of network traffic classification based on machine learning algorithm, such as high accuracy and easy to be extended, random forest is selected as network traffic classification method, and an improved strategy is proposed to overcome the shortcomings of traditional stochastic forest. Finally, the improved stochastic forest strategy is implemented on OCTEON platform, and a network traffic classification system based on OCTEON multi-core processor is designed and implemented. Finally, the system testing environment is built, and the effectiveness of the system is tested. The main work includes the following aspects: 1. On the basis of fully understanding the traditional stochastic forest algorithm, according to the shortcomings of the single node model training speed and the difference in the performance of the base classifier, In this paper, an improved stochastic forest strategy is put forward from two aspects of parallelism and weighting adjustment. The hardware characteristics of OCTEON are studied, and the operation mode of OCTEON is put forward. Combining the excellent characteristics of synchronous communication mechanism and memory allocation mechanism between cores with the random forest strategy of the machine, an improved stochastic forest model, named as "random forest model", is implemented on the OCTEON platform, and a network traffic classification system based on OCTEON multi-core processor is implemented. The architecture design of the system and the detailed design and implementation of each functional module are introduced in detail. The system testing environment is built, and the classification performance of the system is tested, which verifies the excellent performance of the system in traffic classification.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP332;TP393.06

【参考文献】