Exchange Structure And Scheduling And Bandwidth Allocation Algorithm

This thesis consists of two parts: Part one is research on the large capacity scalable switching fabrics and the related scheduling algorithms. Part two is research on traffic control of Ethernet switch, which is combined with a corporation's project to provide solution schemes.With the rapid development of internet, the requirement of network bandwidth increment also grows dramatically. Since the development and application of the optical transmission DWDM, the main factor that limits network bandwidth increment has never been network transmission link. Switches or routers as network service points have been the bottleneck of network performance. They are not only required to provide high bandwidth based on port, but also to make port upgrade freely. Therefore, the work of design of large capacity scalable switches locates in the key position during the progress of designing next generation network, which has been the attention focus of many researchers and technical experts in this field.At present, the research of scalable switch fabric mainly centralizes in the multi-stage switch fabric that is interconnection of multiple switch units in terms of some rules for constructing more large capacity and more ports switch fabric. Multi-stage switch fabric can be classified into many categories according to the internal connection manner. Due to three stage Clos switch fabric has a good scalability characteristic, it has been the best choice for scalable switch construction. This paper's first section is studying in the switch fabric's scalability property. Subsequently based on the improvement of three stage Clos switch fabric, we have designed the flow based scheduling algorithm. Traditionally the multi-stage fabric's scheduling algorithms derived from single stage fabric's scheduling algorithms whose main thought is matching of graphs and finding of the maximum ports connections. Up to now, there are many classical scheduling algorithms, though they can reach good performance, there are some short can not be ignored, for example the lack of QoS guarantee. On the other hand, the scheduling algorithm based on flow can service flows in terms of their different characteristics, aiming at the current network traffic's diversity, it provides QoS guarantee according to flow's requirement, which also is the research hot presently. This paper has constructed simulation model to verify the flow based scheduling algorithm's performance and has provided a feasible scheme of constructing large capacity switch fabrics and scheduling algorithms.This paper's part two is a research of Ethernet switch, which is solution scheme for general problem by Shanghai Opluan corporation put forward: there exists probability of multiple ports persist in sending packet to one common egress port during a continuous time in Ethernet switch, The direct result is occurrence of traffic congestion and losses of packet that affect switch performance badly. Therefore, in order to resolve this problem, we have brought forwards a method that dynamically allocates bandwidth for ingress flows: design an independent scheduler outside of the system that hardware implemented by FPGA. The scheduler communicates information with switch chip through high speed interface and controls ingress flow. The project's goal is to reach avoidance of traffic congestion and provide some degree packet delivery and delay performance guarantee. We have put forward two schemes in terms of scheduling methods: one is scheduling based on egress ports'buffer status; the other is scheduling based on egress ports'bandwidth. At the same time, we have designed two bandwidth allocation algorithms: the one is"fair and reverse algorithm"; the second is"fair and shaping algorithm". The models of these two schemes have been constructed and their performances have been verified. Finally, we have discussed the project's detail problems and given reference schemes.Two parts in this paper have adopted software OPNET to construct simulation models, and have run simulations from many aspects, analyzed simulation results. All of these works have provided scientific verification for engineering practice.