Research On Active Queue Management Algorithms In IP Network

Internet based on TCP/IP has experienced an explosively growth since 1990's. Rapid growth of Internet users, as well as the appearance of varied multimedia applications has brought forth the rapid growth of Internet traffic which cause network congestion. Network congestion has become a significant problem that prevents the development and application of Internet.Internet primarily relies on TCP end-to-end congestion control to avoid network congestion. But TCP strategies couldn't meet many various demands of every application on the complex network. In fact, it will be more effective for detecting and preventing congestion if the routers perform appropriate queue management schemes. Congestion control strategies for the routers have become an active research field. IETF (Internet Engineering Task Force) suggested that active queue management (AQM) schemes should be adopted in routers as a congestion control means of network layer.Active queue Management schemes have aroused tremendous interest and vast study of global scientific research institution and router manufacturers since IETF brought forward AQM technology and recommended RED (Random Early Detection) algorithm. Then, many AQM algorithms have been put forward. But most of AQM algorithms such as RED, etc. lack fairness. They can't allocate bandwidth fairly in competing flows. Therefore, many active queue management algorithms that can support different level of fairness have been proposed recently. CSFQ (Core-Stateless Fair Queuing) is a typical algorithm to achieve fair bandwidth allocation. It distinguishes edge router from core router. Core router needn't to keep per-flow state. CSFQ achieves preferable fairness while at the same time decreases the realization complexity of core router algorithm. But there are still many aspects of CSFQ that need to be improved.In this thesis, firstly, theory analysis and qualitative comparison on AQM algorithms such as RED, FRED (Flow Random Early Detection), DRR (Deficit Round Robin) and CSFQ are made according to their fairness, complexity and restraining effect to TCP flows. Then some AQM algorithms such as DRR, FRED, CSFQ, RED, FIFO (First In First out) are simulated in single and multiple congested link network topologies by using network simulator NS2. The simulation results show that CSFQ performs comparable or even better fairness to FRED while at the same time markedly decreases the implementing complexity. Especially, it outperforms DRR when the flow number is biggish. The results also show that CSFQ restrains TCP flows. DRR behaves best in fairness in most cases. However, the fairness performance of TCP flow decreases evidently when flow number is large. RED and FIFO behave badly in all cases and RED outperforms FIFO clearly.The close relationship between K and CSFQ performance is found out in the thesis by using plentiful simulation experiments. And then a more appropriate range of K is given. The simulation results make it clear that the TCP performance ofCSFQ can be improved by choosing proper K.The thesis brings forward AR-CSFQ algorithm based on CSFQ to solve the problem of CSFQ, such as its restraining effect on TCP flows and buffer management strategy. AR-CSFQ uses ARED as its buffer management algorithm. It is simulated in multiple congested link network topologies. The simulation results show that the improved algorithm decreases its restraining effect to TCP flows and improve fairness to TCP flows while at the same time keep the virtue of original CSFQ algorithm.