Network Performance Analytical Method Based On Fluid Model

Internet has exceeded people’s expectation in its fast development speed and large scale in recent years. Faced with so complicated network, it’s difficult than ever for us to understand its state and behavior. So an effective network performance analysis technology is needed with great urgency, to help those network designers and Telecommunication Operators understand network performance better.This thesis firstly introduces classification of network simulation models, and analyzes the weak points that existed in mostly-often used model, that is packet model. Then congestion avoidance technologies are introduced, including tail drop (TD) and random early drop (RED). Moreover, congestion management technologies including FIFO(First in First out)、PQ(Priority Queuing)、WFQ(Weighted Fair Queuing) are also introduced in this thesis. Then this thesis analyzes the relationship between ripple effect in fluid model and congestion management technologies.After that, two kinds of implementations of fluid model are primarily discussed in this thesis that is time-driven continuous system and event-driven discrete system. In these two systems, the author expands their scheduling policy to support WFQ and PO.Firstly, time-driven discrete system based on fluid model is comprehensively researched in this thesis. The author models TCP flow in the network and Active Queue Management (AQM) in the router, adopting linear differential equation. Then these equations are transported into a set of solvable differential equations, and implemented by C++. After that, the simulation results are compared with OPNET to improve its feasibility and validity.Then, the author analyzes event-driven discrete system based on fluid model. The primary problem for this discrete system model is how to process events occurred in FIFO ports, which are summarized into4mutually exclusive cases. To improve simulation efficiency further, routing protocols such as OSPF and IS-IS can be simplified in fluid model. So the author introduces how to simplify routing protocols.Finally the author establishes a universal network simulation platform called FMSimulator (Fluid Model Simulator), which involves setting up network topology visually, routing, and high-speed network simulation as well. The internal simulation core is based on event-driven discrete system based on fluid model.To verify the correctness of simulation results from FMSimulator, the author designs a small network scenario with6nodes and another big one with26nodes, and have them simulated in FMSimulator. Then simulation results are compared with OPNET in the same scenario. The experiment confirms that FMSimulator can give corresponding results to that from OPNET, and FMSimulator runs3orders of magnitude faster than OPNET does in the same scenario.