Router-based network traffic observation by terminal sliding mode control theory

Xu, L 2018, Router-based network traffic observation by terminal sliding mode control theory, Doctor of Philosophy (PhD), Engineering, RMIT University.


Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Xu_Long.pdf Thesis application/pdf 9.83MB
Title Router-based network traffic observation by terminal sliding mode control theory
Author(s) Xu, L
Year 2018
Abstract Since the early days of the Internet, network traffic monitoring (NTM) has always played a strategic role in understanding and characterizing users’ activities. Nowadays, with the increased complexity of the Internet infrastructure, applications, and services, this role has become more crucial than ever. The aims of NTM are mainly focused on the three improvements, which include the quality of service (QoS) of the network, optimization of resource usage, and enhancement of security in computer networks. Specifically speaking, firstly, network conditions can be recognized by the network manager with NTM scheme. It provides the complete details about the QoS of networks, such as bandwidth, throughput, propagation delay, link availability, jitter, server memory, database space and etc. Secondly, with NTM being implemented at network nodes, i.e., network gateways, such as routers, or network links, the network traffic that is traversing the network is under online observation. Thereby, the network utilization can be improved by optimizing the resource usage to avoid the network congestions. Thirdly, unauthenticated service or approaches to the server will be identified by regularly monitoring the traffic. The network convention and statistics about the traffic will be known easily which helps to troubleshoot the network. Security events will also be investigated and the entry of the user will be maintained for responsibility.

The work in this thesis focuses on the development of an intelligent real-time dynamic router-based network traffic observation (RNTO) by using the terminal sliding-mode theory. The RNTO technique is applied at network gateways, i.e., routers, to estimate the status of the traffic flows at the router level. The aims of the proposed RNTO technique is to estimate the traffic states, such as queue length (QL)in router buffer, average congestion window size (ACwnd), and the queuing dynamics of the additional traffic flows (ATF). The main contributions of the work can be broadly categorized into four parts.

First, the problem of router-based network traffic monitoring is formulated as an observer design by using TSM theory for RNTO applications. The proposed TSM observer in the research is a network-based monitoring, which is implemented into the network gateways, i.e., network routers. Different from the static network traffic monitoring methods, the TSM observer is designed by using control methods based on the fluid-flow mathematical model, which represents the traffic dynamics of the interactions in a set of TCP traffic flows through network routers. By considering the time delay and stochastic properties in the data transmission network, the sliding-mode observation strategy is proposed with its high robustness with system parameter uncertainties as well as the external disturbance rejection. Given the natural weakness of chattering in sliding mode control signal, which can affect the system state, the chattering avoiding technique of the proposed TSM observation was utilized by using a smooth control signal for estimating the abnormal dynamics. It does not need any low-pass filler, which will lead to a phase leg. In addition, for the stochastic dynamics of the network traffics, fast transient convergence at a distance from and within a close range of the equilibrium of the traffic dynamics is essential to quickly capture traffic dynamics in network systems. Thus, a fractional term has been considered in the TSM for faster convergence in system states to efficiently estimate the traffic behaviors.

Second, the issue of internal dynamics in network observation system is studied by proposing a novel full-order TSM strategy to speed up the convergence rate of the estimation error. In the RNTO scheme, the precise estimation for ACwnd is needed to estimate the queuing dynamics of ATF. However, the estimation error for ACwnd is not available and it converges to origin asymptotically, which results in a long response time in estimation. The proposed novel TSM observer has been designed to drive the estimation error for ACwnd to a defined known area in the finite-time, which can be calculated. Thereby, the estimation error of ACwnd can converge to origin asymptotically within the defined area. This strategy has shortened the response time and improves the estimation accuracy. This further improves the estimation accuracy for ATF. The comparative studies are conducted to evaluate the performance.

Third, the issue of algorithm-efficient RNTO is investigated by considering an event triggered sliding-mode observer to reduce the computational load and the communication burden. Instead of the time-driven observation scheme, the control of the sliding mode observer is formulated under the event triggered scheme. The control of the observer is designed to be smooth and is directly applied to estimate the dynamics of the additional traffic flows. The event triggered observation algorithms is developed to reduce the computational load of the network router and the communication resource of output link in the network.

Fourth, the problem of global RNTO is addressed by developing a fuzzy TSM observer by using fuzzy theory to achieve global operation under network uncertainties. The existing RNTO schemes are based on the linearization of a certain network conditions, i.e., a fixed number of TCP connections, which is a constant value N. Given the network suffers from time-varying fading, shadowing and interference and the data rate changes over time, the current methods proposed so far might not effectively and accurately monitor and estimate the traffic dynamics under network uncertainties. The T-S fuzzy models are used to model the traffic dynamics of the time-varying data changes in network link resources, i.e. the time-varying number of TCP sections, N(t) in a mathematical model. Based on the T-S fuzzy models, the fuzzy terminal sliding mode observer is established by using the fuzzy logic theory to estimate the states of the network traffic to achieve the global observation performance under the network uncertainties. In the fuzzy terminal sliding mode observer, the control signal is designed to be continuous for application of estimating the additional traffic flows without the low-pass filter.

To evaluate the proposed RNTO technique, the networking simulator tool Network Simulator II (NS-II) has been used. The proposed RNTO algorithms are coded and implemented into network routers in NS-II. Numerous simulation scenarios are considered and performed. The comparative studies are also conducted by analyzing the NS-2 results. The results have demonstrated the effectiveness and efficiency of the proposed RNTO algorithms.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Engineering
Subjects Control Systems, Robotics and Automation
Computer Communications Networks
Computer System Security
Keyword(s) network-based traffic observation
sliding-mode observers
TCP/IP networks
anomaly detection
DDoS attacks
Versions
Version Filter Type
Access Statistics: 137 Abstract Views, 82 File Downloads  -  Detailed Statistics
Created: Thu, 23 Aug 2018, 13:10:56 EST by Keely Chapman
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us