Study Of Large-Scale Network IP Flows Behavior Characteristics And Measurement Algorithms

As the expanding of networks, study of network traffic characteristics is becoming one of main direction of network performance analysis. It can give essential supporting for some network applications, such as performance forecasting, QoS Service, SLA Service, and even for the network measurement which is the foundation of these applications. Because TCP/IP protocol suites play an overwhelming role in inter-connect networks, study based on IP packets is one of main research directions of network traffic. As the aggregation of IP packets, IP flows can not express the characteristics of IP packets, but reflect the users'behavior in higher layers. And so there are more and more papers studying the influences in network payloads and performances which are given by protocols and users by IP flows analysis. IP flows characteristics are the inherent characters among IP flows in natural networks, and they are also including the relationships among those characters. Several IP flows characteristics such as IP flows length distribution, IP flows rate distribution, and IP flows arrival process are becoming one of most important directions replacing the status of IP packets research.There are two main components in this dissertation: Firstly, several characteristics of IP flows (including TCP flows, UDP flows, ICMP flows) are analyzed, such as flows length characteristics, flow rate characteristics, and flows arrival characteristics, and so as the relationships among them. Based on these, this dissertation describes their usages in discovering the network informing, and also analyzes the main factors which can cause these characteristics. Secondly, based on the IP flows characteristics and modeling analysis, this dissertation proposes several algorithms to upgrading the network measurement performances and security. The efficiencies and robustness of the algorithms are approved by a series of experiments.In Section 1, this dissertation studies those TRACEs which come from the networks with different time, different areas and different payloads, and analyzes the characteristics of flows length, flows rate and flows arrival process by experimental methods, and also their relationships.In Chapter 2, the same or different flows length distribution characteristics are discovered by analyzing those TRACEs. The causes and interference elements of those characteristics are inferred using TCP/IP protocol suites and users behavior analysis, and the incidence of those characteristics is also gauged at following. Based on those discussions, the empirical model of IP flows distribution for large-scale networks are proposed based on the characteristics analysis, whose precision is better than Pareto Model, and whose complexity is less than Double-Pareto Model. Kolmogorov-Smirnov Goodness-of-Fit Test is employed to inspect the effect of this model and its parameters. And then, this empirical model is contrasted with other distribution models that presented by former researchers, and the same and different characteristics among all of these models are discussed, and so do their causes. The possible tendency of IP flow distribution is forecasted based on those discussions at the end of this dissertation.In Chapter 3, this dissertation models the IP flows rate based on protocol analysis from its three main components: TCP flows, UDP flows and ICMP flows. From the analysis of parameters of those models, the flows rate are found out to be determine by some influence factors. And then, those...