Ad hoc network is a new tendency of data transmission in the future. Because of the convenience and necessity of mobile phone and/or portable computer coupled with wireless data services, TCP/IP has become an important topic for the study in wireless networks. However, there were a few difficulties in data transmission that must be overcome due to ad hoc environments and the characteristics of the IEEE 802.11 protocols. In these protocols, MAC layer is our primary research topic. Based on the RTS/CTS signal of dynamic retransmission [4], this study presented an improvement to solve the problems: (1) media resources was easily robbed due to the RTS/CTS signal competition, (2) data frame would be dropped by the IEEE 802.11 protocol due to too many times collisions. In addition, this study modifies the CWND in TCP layer in accordance with congestion conditions. Sender can transmit data packets to the network, in which resources can be completely utilized without any waste or loss. We also use the related parameters from the IP and TCP header to calculate flow numbers. The calculated parameters are recorded in IP and TCP header, and then instantly forwarded to the receiver via routers. The receiver can forward these parameters back to the sender by using back transmission method. The simulation result shows that the proposed methods can effectively improve TCP performance, such as packet loss rate, and fastly increase the CWND, the buffer utilization, and so forth. Thus, the network can perform more effectively while using the MAC-layer RTS/CTS signal.

The number of Web traffic flows dominates Internet traffic today and most Web interactions are short-lived HTTP connections handled by TCP. Most core Internet routers use Drop Tail queuing which produces bursts of packet drops that contribute to unfair service. This thesis introduces two new active queue management (AQM) algorithms, PISA (PI with Short-lived flows Adjustment) and PIMC (PI with Minimum Cwnd). These AQMs are built on top of the PI (Proportional Integrator). To evaluate the performance of PISA and PIMC, a new simple model of HTTP traffic was developed for the NS-2 simulation. TCP sources inform PISA and PIMC routers of their congestion window by embedding a source hint in the packet header. Using the congestion window, PISA drops packets from short-lived Web flows less than packets from long-lived flows. Using a congestion window, PIMC does not drop a packet when congestion window is below a fixed threshold. This study provides a series of NS-2 experiments to investigate the behavior of PISA and PIMC. The results show fewer drops for both PISA and PIMC that avoids timeouts and increases the rate at which Web objects are sent. PISA and PIMC improve the performance of HTTP flows significantly over PI. PISA performs slightly better than PIMC.

In order to easily send data between two end elements without congestion, methods that suitably control flow of date and evaluate possible overload state are necessary. One such method is to control the data flow directly on the transport layer. This layer offers a range of mechanisms dedicated to deal with this issue. The aim of this paper is divided into three parts. The first part describes the integration of transport layer TCP/IP model, and the ability to process TCP data stream. The second part describes methods to manage congestion, their integration by usage environment. It mainly focuses on methods of TCP Reno and TCP Vegas. Their simulation and analysis on transmission the data stream stream. The third part deals with the analysis in detail of TCP Vegas. Analyzes possible parameters for alpha a beta within the TCP Vegas, and a combination of TCP Vegas and TCP Reno.