Studies On Internet End-to-End Congestion Control

The Internet protocol architecture is based on a connectionless end-to-end packet service using the IP protocol. In this architecture, end-to-end congestion control plays a very important role in keeping the stability of the Internet. There are two primary components in end-to-end congestion control: one is the link algorithm executed by network devices, such as routers or switches; the other is the source algorithm executed by host computers or edge devices. Link algorithms detect congestion and generate feedback information, and in response, source algorithms adjust their sending rate.In this thesis, Internet end-to-end congestion control is studied. The main research works and conclusions are as follows:The design of active queue management algorithm (belonging to link algorithm) is studied from different viewpoints, and 3 different active queue management algorithms are proposed. To solve the problem of slow response in PI algorithm, P2I algorithm is proposed by combining the advantage of proportional controller and PI controller. However, the state transition between proportional controller and PI controller in P2I is coarse. Fuzzy-PI algorithm is proposed by applying fuzzy logic to smooth the state transition. Both P2I and Fuzzy-PI are derived from PI algorithm. They both inherit the weakness of PI algorithm and they can only work well in a small range of network environment with fixed parameters. SAP algorithm is proposed by adding adaptive mechanism to proportional controller.How to extend end-to-end congestion control algorithm to support Quality of Service is studied. Based on SAP algorithm, we propose a new AQM mechanism, Weighted SAP (WSAP) to achieve proportional loss rate differentiation. Compared with RIO and WRED, WSAP is easier to configure and is more scalable. Further more, an architecture for differentiated service is proposed in this paper. In the design of this architecture, we pay more attention to the characteristics of network traffic in the Internet and try to preserve the Internet's original design principles. Fairness is an important issue in the research of end-to-end congestion control. The fairness issue in transmission with TCP is studied. To solve the fairness problem of TCP under the situation of different round trip time and multiple congested gateways, EFC (Explicit Fairness Control) algorithm is proposed. The basic idea is to explicitly improve fairness by adding mechanism at both end systems and gateways. Then the fairness in data transmission in the Internet is studied. Internet relies on end system mechanisms to keep fairness in data transmission. As the number of the users has been very large, this scheme has become vulnerable. It is required to deploy some fairness mechanism at gateways. Fair queueing is designed as a gateway mechanism to achieve fair bandwidth allocation. However, in previously proposed architectures, gateway is required to maintain state on a per flow basis. This requirement is hard to meet for core routers. Hierarchical Fair Queueing (HFQ) architecture is proposed to simplify the deployment of fair queueing. This scheme has no scalability problems. It provides convenience for network pricing and protection against some kinds of DoS attacks.