Congestion control in the Internet with mixed traffic sources and heterogeneous wireless access networks

In the future, more and more applications are running through the Internet from different places, often through wireless access networks with different available bandwidths. That creates two problems in the Internet congestion control: (1) mixed traffic patterns generated by different types of applications; and (2) different available bandwidths for an active session from one wireless access network to the other. The mixed traffic usually contains feedback-controlled traffic (responsive traffic) and bursty traffic (non-responsive traffic). The bursty nature of the nonresponsive traffic often causes queue fluctuation and bursts of packet losses at the bottleneck routers. Meanwhile, moving an active session between different wireless access networks often causes transmission rate disruption due to differences in the available bandwidths. In this dissertation, two new solutions are designed to solve the above issues. First, a Wavelet de-noising scheme is designed to classify responsive and non-responsive traffic in the core network. It is only based on the periodic measurements of incoming traffic volume and packet loss rates, without the need to perform line-speed packet header examination. Hence, it will not be limited by network management policies and security mechanisms (e.g. IPSec) and is scalable to the number of flows in the traffic and the increasing link speed. Second, a smooth multi-path handoff scheme is designed at the end systems for the example application: the real-time streaming application. The proposed scheme provides smooth transition of both the data paths and the congestion control states during the handoff. Optimal usage of the different available bandwidths during the handoff is also exploited to improving the video quality. Our results show promising improvement in application performance through extensive simulation experiments.