Research On Key Technologies Of Fairness Guarantee For Bandwidth Sharing

The key idea of Internet is connectionless-oriented packet switch network. The architecture of Internet can't provide isolation and protection to flows. When the load exceeds the capacity of the network, the performance of Internet would decrease dramatically. To provide fairness guarantee on bandwidth sharing is an important problem. In recent years, with the increasing popularity of Internet and the quick development of networking technology, Internet becomes more and more heterogeneous. The network heterogeneity brings new challenges and pressures to networks in the following issues: 1) The impact of different transmission medium. Wireless link and optic link have different properties from wired link. The traditional TCP congestion control schemes have been designed for wired networks, and suffer from performance degradation when we apply them to these links with different types. 2) The impact of different attributes of applications. There exist flows who are unresponsive to the network congestion level, examples include UDP flows, "greedy"flows and so on. These flows are not sensitive to the network congestion level. It is necessary to deploy efficient bandwidth allocation schemes at the switch nodes, so to provide fair share of network bandwidth between different flows. 3) The impact of different distribution of network capacity. Internet comprising different access links of Ethernet,WLAN,DSL and so on. These different links have different bandwidth. It is an important problem to isolate the effect of users with low accessing rate to those users with high accessing rate. To solve the above problems, this thesis addresses on the fairness guarantee on bandwidth sharing. The content of this thesis includes two issues: (1) research on the fairness guarantee on bandwidth sharing between different flows in one queue; (2) research on the fairness guarantee on bandwidth sharing between flows in different queues. The work in this thesis has been supported by the National Science Foundation of China "Investigation of Wireless Multimedia Techniques based on Multimedia Transmission Property"(No.60202005), university collaborate project "Research on Universal Transmission Platform"and corporation collaborate project "Design and Development of High-speed Router". The contributions of this thesis include: 1. The analysis model of multicast congestion control scheme over wireless links. This thesis analyzes the performance of PGMCC -the window-based adaptive multicast bandwidth sharing scheme -over wireless links. The understanding model for the performance problem of PGMCC is proposed as well as the fully discussion on its origin. For its generality, the model can be used for reference to the analysis of other multicast congestion control schemes over wireless links. 2. Method to measure the available bandwidth. In heterogeneous networks with wireless link, the traditional method to measure the available bandwidth using packet loss would bring unfairness to the competing flows. This thesis proposes a new method to measure the available bandwidth. This method can measure the variation on forward-path delay, and discriminate the random loss from congestion loss accordingly. This method increases the accuracy on the measurement of the available bandwidth. 3. Multicast congestion control protocol based on the network status measurement -EPGMCC. To improve the performance of PGMCC over wireless links, this thesis proposes a new multicast congestion control protocol. This protocol achieves significant throughput improvement over PGMCC. EPGMCC increases the fairness on the bandwidth sharing between the competing flows. Throughput improvement of up to 40% can be demonstrated over typical wireless link with 1% random loss rate. 4. Implementation model of packet fair queuing algorithms based on discrete rates and packet lengths. The sorting complexity of PFQ algorithms is F(N) (N is the number of active flows), and in high-speed networks with huge number of flows, it is hard to implement PFQ algorithms. This thesis proposes an implementation model of packet fair queuing algorithms based on discrete rates and packet lengths. This model groups the HOL packets of each flow according to the Service Interval, thus reduces the sorting complexity to F(G)(G is the number of groups), which is independent to the number of active flows N.