TCP Congestion Avoidance Mechanism Based On Network Measurement Technology

Due to the network users widely addition and the shortage of network resources, the phenomena of the network congestion growing occurs. Therefore, the research about the optimization of congestion control becomes a hot topic. Many researchers have proposed a series of network optimization programs based on available bandwidth measurement and network congestion control technology,which greatly enhanced the network performance.The traditional TCP slow start algorithm is a standard algorithm of network congestion control. The TCP Westwood-V algorithm which was optimized on the traditional TCP slow start algorithm, tremendous enhances the network utilization. However, under the conditions of increasing data stream traffic or relatively small available network bandwidth, this algorithm is prone to some problems, such as the low utilization of network resource, delay increasing, overestimated available bandwidth, the TCP global synchronization, and so on.This thesis brings forward a improvement algorithm of TCP Westwood-V with the active and passive measurement techniques; and it takes the random algorithm to adjust the window of congestion control, reducing the TCP global synchronization. The modified algorithms regards the delay obtained by positive measurement as an estimated parameter for user data packet delay and use the passive measured user data packet's information to audit the data obtained by positive measurement and correct it flexibly so as to realize an efficient network delay measurement for the estimated available bandwidth gained by the original algorithm, with the supplement of positive measurement estimation in a certain time interval to measure and estimate the available bandwidth based on the TCP Westwood-V algorithm. This method has many merits such as independence of protocols, low impact on network traffic, simple measurement. Especially in the presence of network congestion even in the case of packet loss, it has the properties of smaller errors, and more precisely measured bandwidth. The simulation was done by the simulation tool NS2 and the optimized algorithm can more precisely measure the available bandwidth and modify the congestion control threshold, reduce the influence on data storming; it can also obviously reduce the TCP global synchronization under the abrupt large amount of data and it utilizes the bottle-neck bandwidth more efficiently and enhances the network performance.At last, we summarize the work of this thesis, and point out the next step of research orientation.