Research And Implementation Of Fluid-flow Network Simulation

Traditionally, packet-level simulation has been widely used for performance evaluation of computer networks. However, this technique does not scale well as the size and complexity of networks increases. The fast growth of computer networks over the past decade makes this approach computationally expensive, if not infeasible, for truly large scale models. Consequently, efficient simulation techniques for such networking models have become an important issue. In this dissertation, we make a mainly study in fluid simulation and the integrating fluid models with packet simulation.This dissertation first discusses the evolution trends and research status of network simulation techniques, describes the sorts and approaches of system simulation and analyses the characteristic of packet simulation and the advantage and shortage of packet simulation techniques.In this dissertation, we first use Stochastic Differential Equations to model the interactions of a set of TCP flows and Active Queue Management in a single congested router network. Then, we present a scalable fluid models and solutions for large-scale IP networks.The tradeoff in fluid model is the accuracy of desired measures of interest obtained, and it is difficult to extract packet-level information from fluid models. However, in this dissertation, we present a hybrid simulation method that maintains the performance advantage of fluid models while providing detailed packet level information for selected packet traffic flows. We present the implementation within Ns-2 in detail. We design a set of experiments to evaluate performance of hybrid models. Comparisons between the hybrid models and a classical packet simulation show that the hybrid models can accurately obtain the transient behavior of the TCP CWND and average queue lengths, and it can accurately describe the behavior of TCP networks even when the TCP flows simulated by the fluid model compete with TCP flows simulated by a packet source while they share the same bottleneck queue, and the speedup of hybrid simulation depends on the amount of the packet traffic in the hybrid simulation.Finally we expect the study of fluid simulation. As a future work, we will improve the fluid models. Another work direction is to further boost the speed of the hybrid simulation. Furthermore, we can integrate physical network interfaces into hybrid models.