Studies On Modeling And Congestion Control Of Network Traffic

With the increasing popularity of internet technologies, it is more thanimportant to study the behaviors of network traffic so as to strengthen theability of network traffic control and improve the performance of networkservice. The thesis pays close attention to this challenging issue, proposingseveral novel approaches which cover modeling of network traffic,estimation of dynamic bursty nature, prediction of network traffic, as well asactive congestion control of network traffic. The details are presented asfollows:1. In roder to interpret the emergence of multifractal nature of networktraffic, a structural ON/OFF model characterized by three layers is proposed.According to the philosophy that network traffic generates from top tobottom along the protocol stacks, each layer of the model represents physicalmeanings corresponding to the statistics of real traffic, while the synthetictraffic can be of the same multifractal characteristics with the real one. 2. In order to describe the bursty nature of network traffic, enablingdynamic and asymptotic algorithms employed to estimate the Hurst exponentvalues and multifractal characteristics respectively are introduced. Theformer can effectively judge the burst characteristics of network traffic inglobal time domain, while the latter can explain the causes of variability ofthe Hurst exponent.3. In response to the limited accuracy of conventional multi-stepprediction of network traffic, the thesis proposes a novel algorithm formulti-step prediction of network traffic based on cascade tree structureswhich could avoid errors stemmed from random distributions. Additionally,this algorithm enjoys good performance with increased prediction steps.4. To improve control performance of queue length of routers, anadaptive AQM-PI controller is presented along with parameter tuningalgorithms. The adaptive algorithm adjusts the parameters accommodating tochanges of network loads, and the tuning algorithm employs the recursivebisection searching approaches to achieve an optimum damping ratio,thereby obtaining the best parameters of PI controllers. So, the queue lengthof router can be controlled near the expected value, and the queue jitter isreduced to a minimum with higher accuracy and stability.