Joint TCP Congestion Control and Wireless-Link Scheduling for Mobile Internet Applications

The Transmission Control Protocol (TCP) is one of the core protocols of the Internet protocol suite, which is used by major Internet applications such as World Wide Web, email, remote administration and file transfer. TCP implements scalable and distributed end-to-end congestion control algorithms to share network resources among competing users. TCP was originally designed primarily for wired networks, and it has performed remarkably well as the Internet scaled up by six orders of magnitude in the past decade. However, many studies have shown that the unmodified standard TCP performs poorly in networks with large bandwidth-delay products and/or lossy wireless links. In this thesis, we analyze the problems TCP exhibits in the wireless communication environment, and develop joint TCP congestion control and wireless-link scheduling schemes for mobile applications. We show that the optimal TCP congestion control and link scheduling scheme amounts to window-control oriented implicit primal-dual solvers for underlying network utility maximization. Based on this idea, we develop joint TCP congestion control and wireless-link scheduling schemes for mobile applications over In ternet with centralized and distributed (multi-hop) wireless links. Adopting queueing delay as the congestion measurement, we construct a class of window-based QUeueIng-Control (QUIC) TCP algorithms for flow congestion control. For the system with a centralized access point, we propose that the access point adopts a queueing-delay based MaxWeighttype scheduler for wireless links. For ad-hoc wireless links, each link-transmitter employs the proposed scheme to control its mean back-off time. Under these scheduling strategies, we show that the ( both centralized and ad-hoc) wireless links are coupled in a desired manner such that QUIC-TCP congestion control and the proposed schedulers entail an implicit primal-dual solver to the intended optimization problem, for which global convergence to optimal network equilibrium can be proven. Different from the existing solutions, the proposed schemes can be asynchronously implemented without message passing among network nodes; thus they are readily deployable with current infrastructure. Moreover, global convergence/stability of the proposed schemes to optimal equilibrium is established using the Lyapunov method in the network fluid model. Simulation results are provided to evaluate the proposed schemes in practical networks.