The Implement And Performance Analysis Of New Congestion Control Protocols Under Linux Environment

New Internet aplications, like facebook and twitter, that can promote communication and interaction between people, have developed rapidly in the last several years. Through the mobile Internet, people can use these applications more conveniently. Mobile Internet is an inevitable trend of the development of Internet, and wireless network technology is the cornerstone of the mobile Internet. The transport layer plays an important role in the network, and TCP is the most widely used transport layer protocol. Therefore, the study on the TCP performance in wireless network is significant.Congestion control is an essential mechanism in TCP. Packet loss is considered as a sign of congestion in wired networks. However, in wireless networks, the channel bit rate error is much higher than that in wired networks. Many packet losses may be due to the channel error. This causes a decline in TCP performance.Many new protocols have been proposed to resolve this problem. These protocols can be mainly divided into two categories: end-to-end and middle-layer-assisted. In this paper, we study three new end-to-end congestion control protocols. TCP_SPC protocol introduces statistic process control (SPC) into the congestion control. TCP_AW protocol is the latest version of the TCP_Westwood protocol, which uses the estimated available bandwidth to set the size of the congestion window. TCP_RACC protocol calculates the available bandwidth at the receiver.We implement the above three protocols in linux kernel, and make some changes to the control strategy of the TCP_SPC protocol to improve its friendliness to the TCP_Reno protocol. This paper describes how these protocols are implemented. Because the first two protocols are sender-centric, in line with the kernel congestion control module interface, we implement them as modules. However, the third protocol is receiver-centric, we achieve it by modifying the kernel source codes directly. Through a number of experiments under several scenarios, we observe that thesethree new protocols can achieve higher throughput than the TCP_Reno protocol, when the network random packet loss reaches a certain level. They also keep good friendliness with the TCP_Reno protocol. So the three protocols can resolve the problem of TCP performance degradation caused by random packet loss to some extent.