TCP Congestion Control Analysis And Mechanisms For Maximizing Network Performance

TCP steady-state Performance is affected by the congestion in the network and toselect an appropriate data for the available capacity (bottle-neck link) is an open issue.This congestion is mainly arises when a large amount of flow (FTP transaction) is to besent. The aim of this thesis is to explain and solve this issue up to the maximum levelusing simulation (NS-2) approach in already defined algorithms. Our measurement isable to predict more accurately TCP send rate to avoid packet loss, increasingthroughput and saw-tooth effect in cwnd while comparing with TCP Reno andNewReno. The simulation result shows that the proposed schema can achieve higherthroughput and lower delay, full and always link utilization with minor packet losswhich will be controlled by using some TCP Reno/New Reno already definedmechanism and modifying slow start and congestion avoidance algorithms.The demand of Internet users for diversified services has increased due to the rapiddevelopment of the Internet, and applications such as video streaming and VOIP thatrequire a QoS guarantee have rapidly become popular. There are a number ofnetwork-layer technologies, such as IntServ and DiffServ, and UDP-basedapplication-layer approaches that provide such high-quality network services over theInternet. However, we believe that these approaches have deployment difficulties due tomany problems. I can guarantee that a transport-layer approach is the best solution forproviding high-quality data transmission services, since TCP controls the datatransmission rate according to the network condition.In this thesis, a congestion control mechanism for TCP is proposed to providepredictable throughput for upper-layer applications. Although we cannot absolutelyguarantee the throughput, we can provide the throughput required by an upper-layerapplication with high probability when the network congestion level is not extremely high.Better performance has achieved by modifying cwnd size of a TCP connection in thecongestion avoidance phase by using information on the available bandwidth of the networkpath obtained by Inline Measurement TCP (ImTCP). In the meantime, an additionalmechanism for guaranteeing predictable throughput for multiple TCP connectionsparallelized in the network is also proposed. We have also proposed a congestionavoidance algorithm using Monte Carlo and Bayesian analysis on heterogeneous networks.Resource management and admission control have been notified to bursty traffic. We present the evaluation results for the proposed mechanism obtained in simulationexperiments and implementation experiments in experimental networks. As a result, weconfirm from all of the experiments that the proposed mechanism can achieve acceptableperformance for the bottleneck link capacity even when much competing traffic exists inthe network and there is almost no residual bandwidth.