Congestion Control Research In Network Layer About Adaptive Flows And Non-adaptive Flows

Internet has been developed for dozens of years, it has been an indispensable infrastructure of today's daily life and it has played an important role in many aspects like information interchange, resource sharing and so on. But the soaring data traffic and the quick change of network application make more demands of internet technology.IP network exchanges packets in a connectionless and end-to-end manner, serve for customers with a"best effort"model. It has simple design patterns and strong expansibility. But internet is complex and heterogeneous, there are competitive relations among different flows while sharing network resources, so if network resources are not enough, network congestion will turn up. And we should control the congestion well, otherwise it will lead to continuous decline of network throughput, even congestion collapse. So congestion control is the key factor of internet robustness, also is the basis of various network application and control mechanism.The fairness of the bandwidth allocation is an important index of the performance of algorithm. Traditional routers make no distinction among packets belong to different flows while transmitting and dropping packets, so the bandwidth of each flow is decided by the amount of data sent by end nodes, this will lead to the unfairness of bandwidth allocation among different flows. AQM that takes no account of fairness can't improve utilization rate of network resources in practical applications. Non-adaptive flows will take up all the bandwidth by its high speed and the weak adaptive flows can't transport packets normally, so the overall utilization of network resources declines rapidly.This paper focuses on several kinds of fair AQM algorithms such as CHOKe, MCHOKe, A-CHOKe, XCHOKe and some other improved algorithms based on research on RED algorithm. It is found that these algorithms still have some defects in fairness. So firstly we propose an improved and stateless algorithm named HCHOKe, HCHOKe picks the packets from the queue self-adaptive, then distinguishes the non-adaptive flows more exact. So it can punish non-adaptive effectively. We carried out some simulation experiments on NS2 to verify its fairness, the results show that HCHOKe perform better in fairness than other stateless AQM; then we improve XCHOKe from some important aspects, and combine with self-adaptive congestion mechanism of HCHOKe, then propose an advanced algorithm called H-XCHOKe, it is a new algorithm with self-adaptive congestion mechanism based on CHOKe hit histories; finally we also verify H-XCHOKe's fairness by the simulation experiments on NS2, results show that H-XCHOKe performs better than XCHOKe, HCHOKe, CHOKe and so on in punishing non-adaptive flows and allocating bandwidth fairly.