Congestion Control Based Web Optimization

The World Wide Web has been created for more than20years old. The primary thought was to creat a network based system that allows for sharing information between scientists all over the world. As it turned out the Web did not only stick to this thought and has assumed massive proportions beyond that. With the diversification of Web content, people has become more and more depedent on it and the expection also become higher. Users not only need to get informations from the site, but also wish to get a good experience.Hypertext Transfer Protocol(HTTP) is running on top of transport layer protocol TCP. TCP is a connection-oriented, end-to-end reliable protocol provides for communation between pairs of end hosts. It has flow control mechanisms, congestion control mechanisms which is suitable for the transmission of large amounts of files. But it also costs a lot to start and finish a TCP connection. The majority of connections on the Web are short-lived and Web transactions often bursted. TCP obviously show poor performance on this kind of application.Due to the existing TCP is not well adapted to the features of the Web transaction, research on the improvement of TCP and HTTP performance has been doing for a long time and will continue. In this thesis firstly introduce concept of congestion control, and technologies used to avoid congestion. Then research on the technologies related to HTTP. Analysis and tested the performance of the SPDY protocol which is proposed by Google company aimed for replacing HTTP. Discover a phenomenon by analysising SPDY protocol. Defined a concept of congestion window constant phase and proposed a algorithm on how to choose the size of the congestion window after a idle moment. Then use ns2simulation to test and verify the performance of the algorithm. At last discussed the benfit of using large initial congestion windows.Run a lot of simulation to compare the performance of different size of initial congestion windows and proposed a reconmmend value.