Research On TCP Over Wireless Ad Hoc Networks

This dissertation mainly focuses on research of TCP over wireless ad hoc networks.We modeled TCP as a Markov chain to analyze the performance of TCP and proposed threenew TCP algorithms to improve the throughput of TCP in wireless ad hoc networks.In the first study, the TCP over wireless ad hoc networks is modeled through joint TCPand MAC analysis. The window variation of TCP is modeled as a connective Markov chain,whose transition probabilities are determined by the packet drop probabilities on differentcongestion windows. Through analyzing the IEEE802.11MAC protocol, we derive thesteady packet drop probability, the round trip time, and the average end-to-end throughputon each congestion window. The total throughput is expressed as the expectation of thethroughput on different congestion windows. Simulation results show that this model wellmatches the throughput of TCP over the chain topology with different number of hops.Secondly, the adaptive fractional window increment algorithm is proposed. The fixedfractional window increment (FeW) algorithm proposed in[1] has an obvious shortage:when the congestion window size is smaller than the bandwidth delay product (BDP), fixedwindow increment wastes bandwidth; while when the window size is larger than the BDP,fixed window increment is prone to result in timeout events. Thus, the adaptive fractionalwindow increment (AFW) algorithm is proposed in this thesis, whose window increment islarger than that of FeW when the window size is smaller than the BDP, and smaller than thatof FeW when the window size is larger than the BDP. AFW achieves the tradeoff betweenthe bandwidth wastes caused by small window increment and frequent packet losses causedby large window increment, and obtains better performance than FeW. Simulation resultsshow that the throughput of AFW is about5%higher than that of FeW.The third study concerns the improved algorithm of TCP-Vegas for wireless ad hoc net-works, named as Vegas-W here, is proposed. Most research considers only TCP-Newreno,while another widely used TCP variation, TCP-Vegas, does not receive enough attention. However, some references show that the performance of TCP-Vegas in ad hoc networks isbetter than that of TCP-Newreno. Since the mechanisms of these two TCP variations arenot the same, we have to consider the specific features of TCP-Vegas. Thus, we analyzethe influence of the congestion window variation of TCP-Vegas on its throughput with TCPwindow model under the unified network model. And based on that, the improved algorithmVegas-W is proposed to improve the throughput of TCP-Vegas in wireless ad hoc networks.The improvement of Vegas-W includes four aspects: fractional congestion window exten-sion, slower start, slow congestion avoidance and slow start threshold update. Simulationresults show that, the performance of Vegas-W is better than that of both TCP-Vegas andFeW algorithm proposed for TCP-Newreno.Finally, the incorporated ACK is proposed to improve the performance of TCP in IEEE802.11based ad hoc networks. Most researches on TCP in wireless ad hoc networks focuseon controlling the sending process of TCP to improve the performance, while the influenceof ACK packets is ignored. However, in IEEE802.11based ad hoc networks, the transmis-sion time of an ACK packet has serious impacts on the performance of TCP. In a typicalscenario, in which the packet length is1024bytes and the transmission rate is2Mbps, thetransmission time of an ACK packet is up to33.8%of that of a DATA packet. However, themajor useful information in an ACK packet is only the destination address and the sequencenumber of latest received DATA packet. The large overhead for transmission of an ACKpacket costs too much bandwidth. Thus, the incorporated ACK is proposed in this article,which incorporates the useful information in end-to-end acknowledgements at the trans-port layer in the CTS packets at MAC layer to save bandwidth. To realize this, we makenecessary modifications of the mechanisms at all the transport, routing and MAC layers.The IACK algorithm is implemented in ns-2and its performance is evaluated extensively.With no ad hoc routing protocol NOAH[2], the throughput and goodput improvement ofIACK compared with TCP-AP is larger than8.0%,8.3%, respectively. And it is larger than6.8%,7.6%when compared with TCP-Newreno. With dynamic source routing (DSR)[3],the throughput and goodput improvement of IACK compared with TCP-AP is larger than10.6%,8.8%, respectively. And it is larger than8.8%,11.7%when compared with TCP-Newreno. When there is error in the wireless channel, the performance improvement issimilar.