Research On TCP-Friendly Congestion Control And Its Synthesis With AQM

As the rapid growth of Internet in size and applications, the majority of users and researchers have paid close attention to network congestion problem. The increasing open-loop applications make congestion even worse, especially with the emerging of multimedia applications that account for a large proportion of the Internet and the increasing multimedia information, which need real-time processing and end to end quality of service. But Internet based on best effort can not guarantee the quality of real-time application service. TCP as the dominant network transport protocol uses congestion control mechanism. The frequent fluctuations of the sending rate in the process of detecting idle bandwidth and responsing to congestion pose a threat to the stability of quality of multimedia application service. Therefore, the real-time data stream transmission is not suitable for using TCP but mostly uses UDP. Since there is no end to end flow control of UDP stream, UDP will not change the sending rate regardless of whether the network congestion occurs. Finally, the network bandwidth is seized by UDP stream severely when congestion occurs, resulting in unfair distribution of network resources, even congestion collapse. The fundamental solution can only be to improve the congestion control algorithm properly, so that the sending rate of TCP flow tends to smooth, and thus suitable for the delivery of UDP streams. That is the key idea of the thesis.On the other hand, now the TCP congestion control algorithms and active queue management algorithm are mostly researched in isolation. In the most case with the improvement of the performance of a certain aspect, the other properties are lower. Thus the study of integrated performance of the network is particularly important. Most available studies are based on the steps of heuristic thinking, direct inference and experimental proof, lack of systems analysis. In the thesis, the mathematical analysis model is used so as to strengthens the theoretical significance of the study.Firstly, this thesis studies the widely used class of network transmission control protocol, namely TCP flow sending rate model. Based on the studies of Padhye's model, this thesis widened the original model assumptions and joined consideration of slow start working condition that ignored by previous studies, an improved TCP sending rate formula called A-P model was derived. The formula's accuracy was tested by measured data and network simulations. Compared with Padhye's model which is more accurate, the A-P model that consider the cases under various working conditions is more theoretical and valuable when packet loss rate is low.Secondly, the core of TCP congestion control algorithm, AIMD, is improved by AAIMD (Advanced Additive Increase Multiplicative Decrease) algorithm. Based on the derivation of A-P model, AAIMD flow sending rate model is proposed. After analyzing the principle of setting parameters, this thesis sets the parameter requirements meet TCP-friendly. Then the smoothness, TCP-friendliness and responsiveness of AAIMD flow are analyzed in different parameters. The results of the simulations show us which AAIMD flow has the best performance.Finally, the thesis adds active queue management algorithm, RED, to the simulation process of TCP congestion control in order to analyze its advantage of upgrading network performance. Then the prediction accuracy of A-P model is verified and the impact on network performance is compared between AAIMD and AIMD. The results of these studies will lay a solid foundation for future applications and researches.