基于混合模式的网络流量优化

发布时间：2018-06-23 04:33

本文选题：流量分类 + 混合模式　；参考：《苏州大学》2014年硕士论文

【摘要】：网络流量分类是指将混合有各种应用的流量按应用协议来进行分类，即鉴别网络报文分组的应用类别的过程。网络流量分类技术除了能够为运营商提供更好的网络服务以外，还能有效地进行监督和管理网络。因此，网络流量分类在优化网络带宽、提高网络服务质量、对特定的应用进行计费、对恶意流量进行监测以及确保网络安全等方面有着极其重要的意义。流量分类的过程主要包括两个步骤：首先是选择适当的网络流属性集，作为分类器所用数据集；其次，选择适当的学习方法对网络流量进行分类。因而选择普适性的特征集和合适的学习方法对于流量分类的结果至关重要。目前，现有的特征选择技术使得特征集过度依赖于样本空间，对于不同网络环境的普适性较低，，而当前研究较多的学习方法是机器学习方法，但其计算效率较低，难以实现实时分类。因此，本文针对上述问题，基于混合模式对网络流量分类方法进行了优化。研究工作主要包括三个方面：第一，本文从量纲分析法的角度对流统计特征进行了规约化，并推导出一组普适性流量特征集；第二，由于深度数据包检测技术相对于统计学习方法来说更为准确、高效、移植性好，是目前商用流量分类系统的主要技术选择，但无法适用于加密流量，而机器学习方法能解决这个问题。因此，本文采用的学习方法是机器学习和深度数据包检测技术的混合技术（即混合模式）；第三，构建了一个分布式平台。通过该平台利用混合方法对流量进行检测和分类处理，利用多个集群系统进行并行处理、监督和调度，以达到平摊资源、避免系统资源崩溃的效果，并且对负载均衡算法进行了改进，从而实现了资源的综合利用，提高了优化效率。实验结果表明，在高带宽骨干网现网复杂流量类别中，本文推导出的规约化方法具有一定的普适性，用于分类的时间较短，效率更高。同时本文在该局域网环境下，利用规约化的数据集进行基于混合模式的分类实验，并分别与以往的这种混合技术以及单一的深度数据包检测技术进行对比，结果表明基于混合模式的网络流量分类平台规模较小，分类速度相对更快，从而进一步说明了系统优化的效果。
[Abstract]:Network traffic classification refers to the process of classifying traffic mixed with various applications according to application protocols, that is, to identify the application categories of network packet packets. Network traffic classification technology not only can provide better network services for operators, but also can effectively supervise and manage the network. Therefore, network traffic classification is of great significance in optimizing network bandwidth, improving network quality of service, charging specific applications, monitoring malicious traffic and ensuring network security. The process of traffic classification mainly includes two steps: first, selecting the appropriate attribute set of network flow as the data set used by classifier; secondly, selecting the appropriate learning method to classify network traffic. Therefore, the selection of universal feature sets and appropriate learning methods is very important to the result of traffic classification. At present, the existing feature selection technology makes the feature set excessively dependent on the sample space, and the universality of different network environments is low. However, machine learning is the most widely studied learning method, but its computational efficiency is low. It is difficult to realize real-time classification. Therefore, in order to solve the above problems, this paper optimizes the network traffic classification method based on hybrid mode. The research work mainly includes three aspects: first, this paper normalizes the characteristics of convection statistics from the perspective of dimensional analysis, and deduces a set of universal flow characteristic sets; second, Since the depth packet detection technology is more accurate, efficient and portable than the statistical learning method, it is the main technical choice of the current commercial traffic classification system, but it can not be applied to encrypted traffic. Machine learning can solve this problem. Therefore, the learning method adopted in this paper is a hybrid technology of machine learning and depth packet detection (i.e. hybrid mode). Thirdly, a distributed platform is constructed. The platform uses hybrid method to detect and classify the traffic, and uses multiple cluster systems to process, supervise and schedule in parallel, so as to achieve the effect of sharing resources and avoiding the collapse of system resources. The load balancing algorithm is improved to realize the comprehensive utilization of resources and improve the efficiency of optimization. The experimental results show that in the complex traffic classes of high-bandwidth backbone networks, the proposed regularization method has a certain universality, shorter time for classification and higher efficiency. At the same time, in this LAN environment, the canonical data set is used to carry out the classification experiment based on hybrid mode, and compared with the previous hybrid technology and the single depth data packet detection technology. The results show that the network traffic classification platform based on hybrid mode is smaller in scale and faster in classification speed, which further explains the effect of system optimization.
【学位授予单位】：苏州大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【参考文献】