Achieving Scalability and Self-Adaptivity to Network Bandwidth and Delay for Measurement-based TCP Congestion Control

The heterogeneity and complexity of the Internet are increasing with the rapid development of such networking technology as wireless and DSL/FTTH access network links, gigabit/terabit-level high-speed backbone network links, and the explosive growth of the Internet population. As network link bandwidth increases, server and client machines directly connect to higher-speed networks to deliver gigabyte/terabyte data to other hosts. When TCP Reno, which is the de facto standard transport layer protocol in the current Internet, is used for such high-speed data trans-mission, it cannot achieve sufficient throughput because of the nature of its essential congestion control mechanism. The main reason is that TCP Reno does not utilize the bandwidth information of the network path, one of the most important factors for congestion control in the Internet. The variants of TCP Reno previously proposed for large bandwidth and long delay networks have the same shortcomings. In this thesis, we propose a new congestion control mechanism based on the measurement of the network bandwidth information. We adopt inline network measurement since it can measure physical and available bandwidths from data/ACK packets transmitted by active TCP connection