Research On Hybrid Packet/Fluid Model On Network Simulation

With the fast development of Internet scale and the increase of network applications, the study of behavior and performance of Network becomes a task that is difficult and important. Network simulation plays an important role in studying network behavior and performance as a tool. Traditional network simulators trade the change of a packet's state as a event, so it is very accurate, but has a high computational cost. Current network scale is far beyond the ability of traditional network simulator. Packet-based simulation is facing a big challenge, and it needs improving. This paper describes many kinds of simulation methods, and improves an implement of one method to achieve a goal better and more efficient.Firstly, current research situation of network simulation is introduced in this paper, including the point of research on network simulation and the popular network simulators at present. A brief illustration of network simulation methods is then present from the view of abstraction, including traditional packet-based simulation, fluid-based simulation and hybrid simulation combining packet and fluid model. Packet-based simulation is accurate but with more computation, while fluid-based simulation is fast but with no detail of packet. Both two kinds of method have their own advantage and disadvantage. Hybrid simulation is a trade-off between packet-based method and fluid-based method, aiming improving the efficiency of simulation while keeping the low accuracy loss. We then analyze a hybrid simulation method called Fluid Flow Model particularly, whose abbreviation is FFM. An enhanced model UFFM is introduced to add UDP flows to original FFM. This model describes the behavior of the foreground packet while background UDP flows and background TCP flows coexist. We implement UFFM with the use of network simulator NS2 in the case of Poisson-distributed arrival rate of UDP flows. A series of experiment is done to verify the accuracy of UFFM under different network environment and the performance of UFFM in large-scale network topology. The conclusion and future work are present at the end of this paper.