Research And Implementation On TCP Congestion Control Mechanism Over Wireless Networks

TCP Protocol is the most widely used protocol in the Internet, which provides reliable, robust end-to-end data transmission service. And the congestion control mechanism in TCP is one of the key techniques to keep the Internet stable. But the original TCP was only designed for a low BER (Bit Error Rate) wireline network, because it assumes that any lost packet is due to congestion rather than transmission error. But wireless telecommunication is characterized by high error rates and low bandwidth compared to wired networks, so when random loss occurs in wireless networks, TCP mistakes the cause for congestion and reduces the transmission rate, leading to low performance. Therefore, how to improve TCP performance under wireless networks is currently a very important topic.This thesis states an Adaptive Bandwidth Estimation Congestion control mechanism (ABEC), which is a sender-side modification of the TCP congestion control meachanism that improves upon the performance of TCP New Reno in wired as well as wireless networks. The key idea is to continuously measure the bandwidth at the TCP sender side via monitoring the rate of returning ACKs. The estimation is then used to compute congestion window and slow start threshold after a congestion episode, that is, after three duplicate acknowledgements or after a timeout. In contrast with TCP NewReno which"blindly"halves the congestion window after three duplicate ACKs, ABEC attempts to select a slow start threshold and a congestion window which are consistent with the effective bandwidth used at the time packet loss is experienced. So the TCP performance under wireless networks is improved.In contrast with the former similar algorithms, such as Westwood, there are three significant improvements. Firstly, we can choose the time interval of bandwidth measurement more appropriately by estimating the cause of packet loss, that is, when the cause is more likely congestion, we prolong the time interval to avoid the estimated bandwidth too aggressively so that the fairness is improved; when the cause is more likely random loss, we shorten the time interval to keep the link utilization from being too low. Secondly, we can choose the bandwidth estimation response rate adaptively based on the current network state. When the network dithering is high, we lower the response rate to keep TCP stable, but when the available bandwidth changed significantly, we increase the response rate to keep the estimation accurately. Thirdly, we can catch the change of route by monitoring the change of round trip time (RTT). Bandwidth estimation algorithms need to get the minimal RTT accurately, but the change of route will make the estimated minimal RTT unaccurate so that the estimated ssthresh and cwnd applied to TCP are unaccurate.ABEC is evaluated under various network environments based on NS2 simulation platform, and compared to many TCP algorithms, such as TCP NewReno, Westwood, and Rate Estimation (RE). The result demonstrates that ABEC not only improves the TCP performance under wireless network, but also guarantees the fairness with TCP NewReno under wired networks. And it also can avoid the negative influence of network dithering , ACK compression and route change.ABEC is then implemented in Linux as a kernel module, with which we can change the TCP congestion control machenism dynamically even though the Linux OS is online. The performance under wireless networks and the fairness with TCP NewReno under wired networks of ABEC are then evaluated in real networks. The result demonstrates that ABEC perform very well under both wireless and wired networks.