基于分布相似度迁移的关键路由设备检测

发布时间：2019-03-18 18:20

【摘要】：基础设施网络(如电力网、互联网等)目前已经成为人类日常生活不可或缺的设施。在基础设施网络设施中,有一些节点处于网络中的关键位置(比如跨城、跨国或跨大洲的转送节点),并且节点上的负载较大。这类关键节点的性能在很大程度上影响或制约网络部分区域的功能。为了提高基础设施网络的性能,需要对网络中的关键节点进行有效利用。然而,关键节点通常是事先未知的,需要在真实网络环境中检测出来。如何检测互联网网络中的关键节点是本文所要解决的问题。本文围绕关键基础设施检测问题,实现了互联网网络行为数据获取系统,并针对所获得的数据,尝试使用机器学习的技术来解决。本文的主要贡献如下：(1)开发了一种在服务器数量多、分布广的条件下的网络行为数据获取系统。该系统采用混合式P2P架构,由一台独立的服务器作为控制台,对整个测量任务进行配置以及监控,其他所有参与测试的服务器组成一个P2P的网络进行测量工作。而考虑到路由负载均衡的影响,系统又采用了Paris-traceroute技术对路径探测模块进行设计。在测量的过程中,针对不同的网络与机房条件,系统设计了不同的通信协议以供选择,大大提高了系统的适用性。(2)提出了一种基于分布相似度迁移的互联网关键路由设备的检测算法,其步骤包括：一,在目标域(当前路由)中通过谱聚类方法自动判断出可疑的路由设备：二,通过分类算法对可疑路由设备进行分类(关键路由设备和非关键路由设备)。对于步骤二,我们提出了基于分布相似度迁移的分类方法。这是由于在真实环境中,不同线路中不同路由设备的行为特征由于若干客观因素(网络状态,路由设备性能等)导致其分布通常不相同。基于分布相似度迁移的方法能够度量路由分布之间地差异,从而可以有效地迁移路由的标记。通过在华为公司提供的真实数据集上进行测试,结果表明本文提出方法能够有效发现线路中的关键路由设备,同时本文方法能够根据不同线路之间的分布相似度迁移来提高分类结果。
[Abstract]:Infrastructure networks (such as power networks, Internet, etc.) have become indispensable facilities for human daily life. In infrastructure network facilities, some nodes are in the key position of the network (such as trans-city, cross-border or trans-continent transfer nodes), and the load on the nodes is heavy. The performance of such key nodes affects or restricts the functions of parts of the network to a great extent. In order to improve the performance of infrastructure network, it is necessary to make effective use of the key nodes in the network. However, the key nodes are usually unknown in advance and need to be detected in real network environment. How to detect the key nodes in the Internet network is the problem to be solved in this paper. This paper focuses on the key infrastructure detection problem, and implements the Internet network behavior data acquisition system, and attempts to use machine learning technology to solve the acquired data. The main contributions of this paper are as follows: (1) A network behavior data acquisition system with a large number of servers and a wide distribution is developed. The system is composed of a hybrid P2P architecture and a stand-alone server as the console to configure and monitor the whole measurement task. All the other servers involved in the test form a P2P network to carry out the measurement work. Considering the influence of routing load balancing, the system adopts Paris-traceroute technology to design the path detection module. In the process of measurement, different communication protocols are designed for different network and computer room conditions. The applicability of the system is greatly improved. (2) A detection algorithm for Internet key routing equipment based on distributed similarity migration is proposed, the steps of which are as follows: 1. In the target domain (current route), suspicious routing devices are automatically identified by spectral clustering method. Secondly, suspicious routing devices (critical routing equipment and non-critical routing equipment) are classified by classification algorithm. For step 2, we propose a classification method based on distributed similarity transfer. This is because in the real environment, the behavior characteristics of different routing devices in different lines are usually different due to some objective factors (network status, routing equipment performance, etc.). The method based on distribution similarity migration can measure the difference of route distribution, so it can effectively migrate the label of route. Through testing on the real data set provided by Huawei, the results show that the proposed method can effectively discover the key routing equipment in the line. At the same time, this method can improve the classification results according to the distribution similarity transfer between different lines.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.05

【相似文献】