| Computer networks based on the TCP/IP (Transmission Control Protocol/Internet Protocol) employ TCP congestion control and shortest path routing. However, TCP congestion control can result in under-utilization of link capacity, low session throughput, and unfairness in its throughput performance over impaired links. Conventional shortest path routing can lead to network congestion and under-utilized links due to uneven distribution of traffic in the network.;To address these problems, this thesis proposes multipath congestion control algorithms for data networks, which combine multipath routing with network congestion control. First, an efficient multipath route discovery algorithm is proposed to find multiple paths in the underlying network infrastructure. The multipath route discovery algorithm can find multipath routes with varying degrees of disjointedness. Second, we develop multipath traffic distribution algorithm to alleviate network congestion by exploiting multipath routes. The proposed "congestion-triggered multipath protocol" requires relatively minor upgrades to the existing Internet architecture.;Recently, there have been proposals to introduce explicit rate signaling into the Internet. Explicit rate signaling has the potential to substantially improve network performance, but requires routers that can support signaling on a per-flow basis. Along these lines, we propose an adaptive dynamic rate controller that computes the rate for flows in response to network status (e.g., network congestion, link underutilization) in order to minimize network congestion and fully utilize the link capacity. We evaluate its performance in conjunction with a rate-based transport protocol. |
