Fairness Of TCP/IP

TCP/IP networking provides only one service: Best Effort datagram delivery. However, with the development of, such as, commercialization and the diversity of application requirements, more concrete definition of quality of service is necessary. Traditional bandwidth allocation method and congestion control principle will not satisfy the changing demands.On the assumption that Internet society will continue to support the Best Effort service model in the future, the dissertation believes that most of applications, with appropriate bandwidth overprovisioning and some control over data flow, in addition with their abilities of adaptation to dynamic network conditions, will perform sufficiently well under moderate congestion levels. Therefore we conduct some research based on two principles:First, network must protect flows by enforcing restrictions on resource usage. Second, the network must provide feedbacks to users in time.The dissertation proposes two fair bandwidth allocation algorithms: MBLUE (Modified BLUE) and UFQR (User Fair Queuing with Reservation).1. MBLUE aims at average fairness. It uses packet loss and link utilization history as an estimator of congestion to detect incipient congestion, and to notify relative connections. MBLUE works with an FIFO queue without assumptions about queuing architecture. Through little state information of each flow MBLUE achieves fair bandwidth allocation among flows sharing the bottle-neck.Simulation results demonstrate the fairness of MBLUE. It also validates the effectiveness of two principle mentioned above.2. Based on the concept of user fairness, the dissertation proposes UFQR, which can allocate bandwidth among users according to their payments during congestion.For each packet at the edge of a network, UFQR inserts a label with user's utility, which is the ratio of its real traffic into network and its reservation. It uses average queue length as the indicator of the severity of congestion, and uses utility as main index of packets dropping.A user, whose traffic is higher than the average service level, will receive negative feedbacks from UFQR during congestion, and be notified to reduce his packets sending; whereas a user whose traffic is lower than average, will receive positive feedbacks and be notified to increase his packets sending. If a user ignores negative feedbacks from network, packets out of his fair share will be dropped by UFQR. So with end-to-end efforts, the bandwidth is allocated fairly during congestion.The dissertation analyzes the scheme of user fair bandwidth allocation with game and pricing theory, and shows its rationality and equilibrium.Simulation results are presented to validate the performance of the algorithm under different scenarios.