Improvement Of TCP’s Congestion Control Protocol

Nowadays, computer network technology has been rapidly developed. Networkhas been playing an increasingly important role in our daily life. Internet users increasedramatically and network traffic also increases dramatically. Once the networkcongestion situation happens, network will suffer a large number of packet losses andthroughput will also be influenced, and even serious network collapse will happen.Therefore congestion control for network is of great significance to keep the networkdeveloping.The thesis mainly studies network congestion control protocol based on TCP/IPprotocol. And Reno-J algorithm is proposed to solve the shortcomings of the Renoalgorithm so that Reno-J algorithm can avoid the emergence of a large number of burstdata during the slow start stage:Firstly, the thesis analysed the TCP protocol in detail. The thesis mainly introducedTCP segment header format. Through detailed analysis of three-step handshake andsliding window mechanism, it can be illustrated that how TCP realizes reliabletransmission. The thesis respectively explains the causes of congestion and congestioncontrol concepts and analyses the slow start, congestion avoidance, fastretransmission/fast recovery, timeout retransmission working principle. The thesisbriefly expounds the evolution of congestion control protocol version.Secondly, comparative analysis is mainly focused on the advantages anddisadvantages of the Westwood algorithm and the upgraded versions and improvementis made on bandwidth estimation algorithm according to the problem that currentcongestion control algorithm sets ssthresh constantly. The improvement is improved bysampling the ACK packet length and ACK packet’s arrival time for getting their averagevalue respectively, and the estimated bandwidth is obtained by dividing the averageACK packet length by the average ACK packet’s arrival time. Improved availablebandwidth estimation algorithm can estimate the network bandwidth more accurately.Finally, TCP Reno increases its congestion window exponentially during the initialslow start stage, this will lead to the emergence of a large number of burst data at the late stage of slow start. For this reason, a piecewise mathematical function model whichis symmetrical about midpoint is proposed. The congestion window growth rate isaccelerated at first and the congestion window growth rate slows down at last duringslow start stage so that it can smoothly switch to the congestion avoidance stage. Itscongestion window, throughput, packet loss rate, friendliness are evaluated andvalidated by NS2simulation. The experimental results show that the new algorithm caneffectively reduce the losses for network transmission of packets and improve thethroughput of network. The new algorithm can also decrease the number of networkjitter and maintain network stability. In addition, when the new algorithm and Renoalgorithm are applied together, the new algorithm is friendly to Reno algorithm.