Improvement Of Tcp Based On Loss Differentiation. Wired - Wireless Network

TCP is designed based on wired network, and considers all losses as congestion signals. But it often goes with wireless losses in wireless circu mstance, so triggering congestion control will reduce the sender's congestion window. The aim of the loss-differential scheme is to disti nguish congestion losses from wireless losses, by this way avoiding start-uping congestion control blindly and improving the performance of network.The paper firstly discusses the related information of the Throughput, and finds out the reasons for poor performance in wireless network. This can be attributed to the lack of loss-differential and relevant disposal, then we summarized and analysed some other ideas of the loss-differential schemes, and concluded that ABQM scheme can respond to the change of network states, based on ABQM scheme, the paper improves the scheme from the following aspects:Firstly in order to ensure only one packet marked out in a marked -interval, we gain the self-adaptive adjustment mechanism which adopts AIMD (1, a) to adjust the marked-interval, the variable a changes according to the number of packets marked out in continuous three losses, and makes marked-interval responds to actual network states.Secondly for continuous losses or single loss we propose the proximity principle algorithm to differentiate them or it. by seqno of the lost packet together with bias , marked-interval and seqno of the biggest packet which is seen before the lost packet, the receiver can calculate bias of the lost packet.Once more we propose the assigned way of the marked-interval in the sender. In order to guarantee the precision of loss-differential scheme, the paper comfirms the range of marked-interval and the maximal value of marked-interval equals to the value that the bottleneck link can accommodate the packets in different circumstance.Thereout, the paper attains A+BQM scheme. The experimental result shows A+ BQM scheme not only improves the network's performance, but also has higher accuracy, robustness and fairness in any circumstance.