Research On Explicit Congestion Control For High Speed Networks

Congestion control which is a critical issue in computer networks is one of the keytechniques for the stability of Internet. With the rapid growth of the backbonebandwidth of next generation Internet and the demands of data transmission of variousnew network applications, explicite congestion control mechanism assisted by routersattracts great attention.There are still some open problems in the research area of explicit congestioncontrol, such as devising algorithms with fast convergence, setting systematic parameter,adaptively, designing robust algorithms in large delay networks, fair bandwidthallocation, and performance evaluation, etc. We propose a series of new algorithms andmethods in this dissertation to solve the problems mentioned above.Convergence to equilibrium fastly is a desired property to evaluate congestioncontrol mechanism. Through detailed analysis, we conclude that constant parametersetting slows the convergence speed of the algorithms. We propose a new algorithmcalled NMKC which adaptively adjusts parameters based on a feedback function. Then,the stability condition of NMKC with any delay is proved. The fairness convergencerate of NMKC is much better than existing algorithms. The simulation results show thatNMKC performs much better in many aspects, such as efficiency, stability, fairness andscalability.Measuring the degrees of congestion and predicting its future trends accurately andquickly are critical issues in congestion control mechanisms. Our research shows thatthe source node can accurately predict the trend of network congestion based onhistorical information. However, most algorithms neglect this important information.We propose a new algorithm called FMKC. Based on historical information, FMKC candetect and predict congestion accurately and fast. We prove theoretical bound of thestability and convergence speed of FMKC. Extensive simulations match the conclusionsof the theoretic analysis perfectly, furthermore show that FMKC ourperforms EMKC inmost situations.The performance of explicit congestion control mechanisms is influenced bygranularity of feedback. We propose a new fair bandwidth estimation algorithm calledATCP. The header of data packet carries congestion information which includes packetloss rate and estimated fair bandwidth. The router uses a heuristic mechanism to detectavailable bandwidth based on the changes of packet loss rate. It is not necessary forrouter to maintain Per-flow state information. The simulation results show that theconvergence time is independent of bandwidth and RTT. And ATCP also has goodperformance such as scalability and fairness.Most of the congestion control mechanisms have difficulty in obtaining desirable delay because the network delay is highly heterogeneous and the behavior of system isvery complex. We present a mechanism called XCP-SVC which exploits somecorrection mechanism to overcome the adverse effects of large delay. The performanceof XCP-SVC is evaluated through NS simulations. The results show that it actuallyimproves the performance of XCP.In summary, we focus on fast convergence, systematic parameter setting, robustalgorithms in large delay networks, fair bandwidth allocationof exolicit congestioncontrol. Our works have academic and practical value for high efficient data delivery inhigh speed networks.