Study On Performance Modeling Of TCP Protocol

With the increasing network application services, especially multimedia applications have appeared, they impose a serious congestion problem in the Internet network. To better control the degree of congestion, there is a requirement to further understand the Internet network and its performance. At present, the understanding and exploring of the Internet network and its performance have the following three methods, namely, the actual measurement, computer simulation and analytical model. The analysis model can be well adapted to the complex network scenarios, relatively accurate performance of the network is estimated.In the past two decades, the models for TCP are becoming more and more popular. By modeling the TCP protocol, we can find the impact of various parameters changes on the TCP performance, discover the defects of TCP protocols, and provide useful instructions for later improvements. In addition, the modeling results of the TCP protocol can be directly applied to the TFRC protocol, which can improve the performance of network transmission. We model TCP performance in some network scenes in this dissertation, and the main work and contributions are as follows:1) Modeling TCP Sack performance in lossy networkMost literatures are to measure the steady state throughput of TCP Reno, however, few existing literatures are on modeling the throughput of TCP Sack. This dissertation provides an analytic model, called TST model, for the TCP Sack steady state throughput. The presented model not only captures the behavior of fast recovery mechanism of Sack which is the main difference with Reno, but also takes the timeout phase into accout which mainly considers the unsuccessful retransmission timeout. Finally, we model the slow start phase and the situation of congestion window under limitation in TST model. Those results simulated by ns2show that TST model can capture steady state TCP congestion window, Sack throughput and situations of congestion window under limitation more accurately than existing TCP models.2) Modeling TCP performance in reordering and lossy networkThere are two types of packet transmission errors in network, packet loss and delay. Different errors will lead to different influence on network performance. This dissertation studies the problem of modeling TCP throughput in packet delay and lossy network, and proposes an analytical model, called TDT model, to predict the throughput of TCP Tahoe in reordering and lossy network. Firstly, we give a sufficient and necessary condition to identify whether a data packet flow is slightly out of order or in order. Based on this condition, the probability that a data packet flow is slightly out of order or in order is analyzed. Then, we consider the probability in reordering and lossy network. Using former research achievements, we can get the TCP maximum congestion window size from the obtained probability of data packet flow is slightly out of order or in order. Finally, TCP throughput as a function of the obtained TCP maximum congestion window size is evaluated. Numerical results indicate that the difference between the probability of data packet flow is slightly out of order or in order of our model and that of real network is tiny, and compared with those existing models which do not consider packet delay, TDT model reduces TCP maximum congestion window evaluation approximately by43%and reduces the TCP throughput evaluation approximately by47%.3) Modeling TCP performance in bi-directional packet loss networkData packet and ACK packet may occur loss when network router buffer overflow. In this dissertation, an analytic model, called TBT model, for the TCP Tahoe performance in bi-directional packet loss network is proposed. This dissertation analyzes the probability of fast retransmission result from single data packet occurs loss in bi-directional packet loss network. For a data packet flow, we first calculate the probability of fast retransmission result form each packet occur loss. Then, Summing of the fast retransmission probability of the packet flow, the probability of data packet flow does not result in fast retransmission is obtained. TCP throughput from the obtained probability of data packet flow does not result in fast retransmission and the performance model is evaluated. Those results simulated by ns2indicate that compared with those existing models TBT model reduces TCP maximum congestion window evaluation approximately by9%and reduces the TCP throughput evaluation approximately by7%.