Research On The Queuing Simulation Of Routers

With the rapid development of Internet, the research of network's related key technologies becomes more and more active. Routers are one of the most important devices in the Internet. The router's switching capacity and throughput has always been the key topics in both industrial and research area. On the one hand, routers must sustain the link rate which grows faster and faster, and on the other hand, they have to offer some kinds of quality of service. The performance of routers is influenced mainly by switching fabric and scheduling algorithms. Current technology development makes it feasible to build extremely high throughput router, which explicit employs switching high-performance fabric. However, the research of scheduling algorithms which have high efficiency and easy implementation is still a challenge. In the practical application, input-queued Crossbar can provide more bandwidth than the traditional share-memory switch. Therefore, this structure is more and more used in high performance routers. The input-queued Crossbar scheduling algorithm has become a research hotpot.First of all, based on the simulation of queuing and the Crossbar Switching fabric, the current input-queued scheduling algorithms are classified and compared from the point of switching model's view, and most studies in this field are summarized. The emerging hotpot of the current research for the new direction and methods are emphasized on the analysis. Secondly, a new algorithm idea is introduced, that is solving the problem of the slot scheduling by the graph theory. An algorithm has been improved by the previously proposed algorithms in graph theory. It is the theoretical foundation to do further research on scheduling algorithms. Thirdly, due to the disadvantage of the algorithms, a serial iterative matching algorithm and parallel iterative matching algorithm which considered the weight are put forward. Simulation model is designed and the simulation tests show these new algorithms have better performance than the previously proposed algorithms. Finally, the next research orientation is discussed.