The Internet has remarkably changed all aspects of our daily life, and while it has yet to reach its full potential, it unfortunately, relies on the TCP/IP architecture which has already shown its limitations. In the layered design of the TCP/IP architecture, the Network Layer only provides best-effort packet delivery service, while the efficiency and fairness are relinquished to the congestion control mechanism implemented in the Transport Layer of the end systems. The TCP congestion control algorithm originally designed for the homogenous wired network is facing great challenges in the ever changing and heterogeneous Internet. For example, conventional TCP cannot perform well in networks with high bandwidth-delay product or in wireless networks with non-congestion-related packet losse...[ Read more ]

The Internet has remarkably changed all aspects of our daily life, and while it has yet to reach its full potential, it unfortunately, relies on the TCP/IP architecture which has already shown its limitations. In the layered design of the TCP/IP architecture, the Network Layer only provides best-effort packet delivery service, while the efficiency and fairness are relinquished to the congestion control mechanism implemented in the Transport Layer of the end systems. The TCP congestion control algorithm originally designed for the homogenous wired network is facing great challenges in the ever changing and heterogeneous Internet. For example, conventional TCP cannot perform well in networks with high bandwidth-delay product or in wireless networks with non-congestion-related packet losses. On the other hand, designed for packet routing, IP networks are vulnerable to misbehaving end systems. Quality of Service (QoS) enhancements on IP networks have been proposed for decades, yet they are still not deployed widely due to their complexity. With the current scale of the Internet, any radical change to the TCP/IP protocol stack is unlikely; simple but effective enhancements involving modifications in only one part of the Internet, either the end systems or the intermediate routers, seem to be more practical.

This thesis focuses on congestion control and fair bandwidth allocation issues in the Internet and discusses how the intermediate routers and the end systems can be enhanced to adapt to the heterogeneous Internet. More specifically, we study the performance of TCP protocols in high-speed networks with the emphasis on how the synchronized loss effect affects the fairness principle achieved by many newly proposed TCP variants in high-speed networks. We also investigate the issue of taming misbehaving flows in IP networks. We propose a new low-cost AQM scheme, SiFTM, to provide better intra- and inter-protocol fairness to support heterogeneous transmission protocols. SiFTM can prevent end users from acting intrusively thus promotes end-to-end congestion control mechanisms. Finally we study the TCP performance in IEEE 802.11 WLAN which is currently the most popular wireless network application in the Internet. We identify some useful techniques that can improve the performance of TCP in such WLAN scenarios.