Congestion Control For High Speed Network Source Algorithms' Research

With the development of computer networks and optical networks, there appearances many high-performance networks with bandwidth larger than G-bit or even T-bit. Some remote control, real-time detection can transfer data and images by high-speed networks. The application of high-speed networks also get widespread increasingly, the study of congestion control mechanism are becoming more and more important. The traditional TCP congestion control mechanism is an end-to-end transmission protocol which be used widespread and has dominant status. It is a important factor to ensure the robustness of Internet. But the traditional TCP with AIMD(Additive Increase and Multiplicative Decrease)strategy can't meet the development requirements of high bandwidth-delay product network. So,more and more scientists begin the research of the TCP congestion control mechanisms for high-speed networks and make it become a hot research field.At present, the study of high-speed networks is still in the initial stage at home and abroad, there have been some representative algorithms: HSTCP, STCP, BICTCP, CUBICTCP, H-TCP etc. These new agreements greatly increase the performance in the high-speed networks through adjusting the increases and decreases mechanism of window, among these, STCP algorithm is a good one, it increase and decrease its congestion window with a constant, and don't need to modify intermediate nodes and receiver nodes which can achieve more easily in a real network. Research shows that STCP can obtain very high throughput and stability in the current high-speed networks, but there are still some serious performance deficiencies. When the queue management adopt the algorithm with drop tail, the stream with small RTT can obtain very large bandwidth resources and the flow with large RTT can be received very little network resources, STCP algorithm is short of RTT fairness; STCP is a typical MIMD(Multiplicative Increase and Multiplicative Decrease)algorithm, which the increase of window is more fast and the decrease is more slow. When the STCP and traditional TCP share the same bandwidth and the packet loss rate is small, STCP will seize a large number of bandwidth than the traditional TCP. So, TCP friendliness is very poor; When several STCP flows begin to run at different times, the first running flow can get greater congestion window because of reducing the window slowly, So that, the new coming flow can be received very little resources and increase window slowly, these flows can't be ultimately converged to a fair and stable status where they get fair bandwidth allocation.The main works in this paper are as follow:(1) Improving the STCP algorithm, the new algorithm was named NSTCP. NSTCP modifies two shortcomings of STCP: bad RTT-fairness and low TCP-friendliness. The increasing window in STCP is inversely proportional to the RTT, so the algorithm is bad RTT-fairness. By adding a fair factor BWE / BWEm ax*RTT , we can eliminat the relation of proportion between the increasing window and RTT, the RTT fairness of the algorithm is enhanced, here, BWE /BWEm axis the current networks bandwidth utilization, which can be more effective to improve network throughput. By estimating the current networks bandwidth, adjusting switch mode between the traditional TCP and NSTCP, avoiding NSTCP flow excessively to occupy network resources and setting aside more resources to traditional TCP, the TCP-friendliness was improved. In NSTCP, two methods are used to make good performances and keep the merits of STCP: high bandwidth utility and stability.(2) Considering bad convergence of the STCP, a convergent STCP congestion control mechanisms was presented in this paper, which was named CSTCP(convergent STCP). The main ideas of the new algorithm are: when a flow'current window is falling, if the flow has larger window than its fair window, we determine to take drastic way to reduce the window or the same slow way to reduce the window with STCP algorithm, this way of the reduction of window will allow the first CSTCP flow to reduce congestion window rapidly and leave more bandwidth resources to the new coming CSTCP flow, in order to achieve equitable share and effective convergence of CSTCP flows. By the network simulation, the new algorithm keeps a high link utilization rate, good RTT-fairness and TCP-friendliness.At last, the above two algorithms NSTCP and CSTCP were integrated to a new unitive algorithm. The basic NS simulation experiment results show that, the new unitive algorithm improves the RTT-fairness,TCP-friendliness and convergence, and keeps the advantages of high bandwidth utilization and stability.