A Research Of Improved Bandwidth Fairness To Different Data Flows In The Internet

1. Introduction With the information society coming, new network applications have continuously appeared. These applications are urgent to the need of QoS, make QoS be hotspot to study. Fairness is one of the important indexes that measure QoS, for the different network applications, it is very essential to guarantee the different flows to obtain fair allotment in the bottleneck link. For solving the above problems, there are some active queue management mechanisms on the router to identify different flows and allot fair bandwidth. These mechanisms can distinguish from different data flows, thereby can more fairly assign bandwidth in the bottleneck link. 2. TCP Throughput Model TCP is the dominant congestion control protocol in the Internet today. The thorough research not only can more effectively carry out TCP, but also can provide the academic foundation and standard for studying of this paper. There are the most typical two kinds of TCP throughput models: simple model and complicated model. simple model Assuming the receiver sends an ACK for every data packet, the maximum sending rate for a TCP connection is thus Tbps, for T—the maximum sending rate; B—the packet size for a TCP connection; p—packet drop rate; tRTT—round trip time. l complicated model Considering the problem of the TCP retransmit timeout, we can get the TCP response function, the formulation is the following: tRTO—retransmit timeout value; b—the receiver receives the number of packets when it returns a ACK, and we suggest that its value is 1. In the process of the TCP congestion control, we can see that a few factors can influence the bandwidth allotment of the TCP flows, mainly including round trip time of packet, the number of microflows in the target converge, target rate size, packet size and non-response (if the UDP flow) existence or not. 3. Improved Research to Bandwidth Fairness of Data Flows 3.1 The Method of Implementing Queue Fairness It is one of the reasons that injure fairness that the different flows have different packet size. Because the IP packet size has very great difference, if all packets carry out uniform drop probability, it will be disadvantageous for the shorter packet of flows, and cause unfair bandwidth allotment. According to the TCP rate response function, EF-RED (Enhanced-Fairness RED) algorithm makes two improvements of the following: l the improvement of final drop probability: EF-RED algorithm considers the influence of each one arriving packet when it is calculating final loss probability, we will change it as: B—the arriving packet size; M—the maximum packet size. l The improved of the parameter countVIII For making EF-RED obey Even Distribute, we will change the calculation of the parameter count as: 3.2 The Method of High Bandwidth Flows Based on Packet Drop History BDH-RED(RED with Based on Drop History)algorithm is the queue management mechanism based on the history of packet drops from a queue with RED. This mechanism can run periodically in the background over information about the flows queuing at the router. Once a active flow appears to drop packet, it will keep this flow state, thus we will judge the connection whether it is high-bandwidth flow. With RED gateways, the number of randomly dropping a packet from a particular connection is roughly proportional to that connection's current arrival rate. In the interval t, we have: aiaiNNRR = (5) Na—the aggregate number of packet drop; Ni—the number of packet drop of the connection i; Ra—the aggregate average arrival rate; Ri—the average arrival rate of the connection i. We adopt the RED packet drop history as the basis of distinguishing the high bandwidth flows in the BDH-RED algorithm. We need to compute the number of packet drop in the history. According to the theorem of Hoeffding and Chernoff-type formula, we can deduce the number of packet drops as follows: The combined drop metric uses the linear combination of the flow's packet drop metric for the random packet drops and the flow's byte drop metric for the forced packet drops in the interval. 2count = count+???MB???(4) [ ]n ≥ln ???? ???c1 ???clbin ???? 1P1m??axcbbii????1?c bi????(6)M (C )= M(R,P)×rR +M(F,B)×rF (7) M(C)—the combined drop metric; M(R,P)—the flow's packet drop metric for the random packet drops; rR —the fraction of the total packet drops from the number that are random; M(F,B)—the flow's byte drop metric for the forced packet drops; rF—the fraction of the total packet drops from the number that are forced. 3.3 Rate-Based TCP-Friendly Unicast Congestion Control When the UDP flows share the same network with the TCP flows, the TCP flows examine to congest, then reduce their sending rate according to congestion control strategy, but the UDP flows continue to send the data with the fixed rate because these applications do not adopt congestion control strategy, these flows result in unfair bandwidth allotment to the TCP flows. For solving the problem, we present the Rate-Based TCP-Friendly Unicast Congestion Control (RTUCC) proposal, the UDP applications increase congestion control mechanism in the algorithm, this algorithm insures the peaceful coexistence of the UDP and TCP data flows. l rate control function We adopt the TCP sending rate function in the complex model as control function, calculate the maximum sending rate of the TCP flows, and simplify the function( we assume b is 1). l parameter estimate Round trip time estimate: we use the method of Exponential Weighted Moving Average: t RTT = (1 ?ω)×tRTTold+ω×tRTTsample (8) tRTTold—round trip time in the last interval; w—disintegration weight; tRTTsample—round trip time in the sample periods. Retransmit timeout estimate: t RTO = 4 ×tRTT (9) Loss event rate estimate: let a loss interval be defined as the number of...