Research On Key Techniques Of Providing Quality-of-Services In High-Performance Router

Advances in communication technologies and inventions in network applications have posed new challenges to performance and functionalities of Internet router. In the last decade, the Internet has evolved into the largest data communication network in the world, and its backbone network's bandwidth and traffic have been increasing explosively, which requires high-performance router to meet the demand of broadband physical link and to realize interconnection. Moreover, with the transformation into a commercial infrastructure, the Internet should provide meaningful QoS (Quality-of-Services) for different applications according to their requirements in the router. The main difficulty in developing QoS technologies for high-performance router is to solve the problem of how to guarantee the high throughput of router while simultaneously satisfy the QoS requirements of packets, such as bandwidth, delay and loss. Therefore, it is necessary to develop new high-performance router architecture and associated QoS algorithms in both data plane and control plane to deal with this new situation. Compared with traditional router with only best effort service, QoS-capable high-performance routers not only utilize switching and parallel processing to improve the system's throughput, but also need support from new QoS scheduling and control algorithms.This thesis explores the key techniques of QoS in high-performance router from three aspects: congestion control, admission control and input-queued scheduling algorithms. The main contributions are given as follows:Congestion control mechanism is one of the key techniques in maintaining the stability of the Internet. In this thesis, a new distributed congestion control scheme, Two-Phase Distributed Congestion Control, is put forward. The scheme divides the end-to-end congestion control into two phases: core network control and access network control. The two phases work independently and can select appropriate control mechanism according to the network's architecture. In the core network control, a novel active queue management algorithm, namely Rate and Queue controlled Approximate Fair Bandwidth allocation (RQ-AFB), is proposed to solve the unfair bandwidth sharing problem between adaptive and non-adaptive flows for the current Internet. RQ-AFB dynamically adjusts dropping probability of an incoming packet based on per-flow's estimated rate, fair allocated bandwidth and queue length. In addition, a new edge-to-edge control scheme, namely Edge-based Aggregation Congestion Control (EB-ACC) is put forward to solve the problem of fairness among aggregates and congestion collapse for Differentiated Services (DiffServ) model. EB-ACC scheme adaptively adjusts sending rate of aggregate at ingress router based on a TCP throughput model and feedback of...