Research Of Queue Management And Admission Control In Network QoS Control

With the evolution of Internet, there has been an increasing demand for Internet to support QoS control to meet various service requirements from different networking applications and to better utilize networking resources. On one hand, end users want to get QoS guarantee when they use Internet; on the other hand, the ISPs want to optimize network resources and achieve more controllable, more manageable Internet. Under this background, network QoS control attracts lots of researchers. QoS control also becomes an important property of the next generation Internet.Indeed, the essence of network QoS control is resource management. So the key techniques of QoS are queue management and admission control. These two aspects are the main contents of this dissertation.In the research of queue scheduling and queue management, we proposed a simple and effective core-stateless virtual clock algorithm, which eliminate statefulness from virtual clock algorithm. This novel core stateless virtual clock algorithm can provide rate guarantee for per-flow and keep good scalability at the same time. In the study of active queue management, we proposed a prediction-based fair active queue management algorithm named PFED. Because of the introduction of traffic prediction and detecting incipient congestion using combined criterion, PFED can achieve good performance on stability and packet drop ratio. At the same time, a stateless method is used to impose effective punishment on misbehaving flow, which improved fairness of this algorithm.Moreover, we studied end-to-end delay bound of packet. We proved that the maximum packet of a flow would never suffer queuing delay when it is forwarded in a fixed rate network. Thus, the end-to-end delay bound of any packets was got. Moreover, this conclusion was generalized to practical packet switching networks, and the end-to-end delay bound in practical networks was got.We also studied traffic measurement, which is the basis of network QoS control. We proposed an adaptive time-driven packet sampling technique based on the study of classic statistic theory and lots of experiment analysis of real-world traffic trace data. This packet sampling can vary sampling degree with the change of network traffic, thus guarantee the error rate under a given level. In this way, high accuracy can be got at low cost. The idea of dynamic changing measurement scheme according to network state can be widely used in network measurement.Finally, we studied admission control. We proposed an end-to-end measurement-based admission control in DiffServ network. Through end-to-end measurement we can got end-to-end available bandwidth, which can be used to make effective admission decision. This admission control scheme can achieve QoS guarantee and high network resource utilization at the same time.