Algorithms and protocols for efficient multicast, transport, and congestion control in wireless networks.

摘要

Effective and efficient support for wireless data transfer is an essential requirement for future Internet design, as the number of wireless network users and devices, and the amount of traffic flowing through these devices have been steadily growing. This dissertation tackles several problems, and proposes algorithmic and protocol design solutions to better provide such support. The first problem is regarding the inefficiency of multicast in wireless networks: a transmission is considered a unicast despite the fact that multiple nearby nodes can receive the transmitted packet. Random network coding (RNC) is considered a cure for this problem, but related wireless network radio resources, such as transmit power, need to be optimally allocated to use RNC to its full advantage. A dynamic radio resource allocation framework for RNC is proposed to maximize multicast throughput. Its efficacy is evaluated through both numerical and event driven simulations. Next, we present the design of MFTP, a clean-slate transport protocol aimed for supporting efficient wireless and mobile content delivery. Current transport protocol of the Internet, TCP, is known to fall short if the end-to-end path involves wireless links where link quality varies drastically, or if the client is mobile. Building on a mobility-centric future Internet architecture, MobilityFirst (MF), a set of transport protocol components are designed to collectively provide robust and efficient data transfer to wireless, or mobile end hosts. These include en-route storage for disconnection, in-network transport service, and hop-by-hop delivery of large chunks of data. A research prototype is built and deployed on ORBIT testbed to evaluate the design. Results from several wireless network use case evaluations, such as large file transfer, web content retrieval, and disconnection services, have shown that the proposed mechanisms achieve significant performance improvement over TCP. Finally, a scalable, network-assisted congestion control algorithm is proposed for the MobilityFirst future Internet architecture. In MobilityFirst, various intelligent functionalities, such as reliability and storage, are placed inside the network to assist with data delivery. Traditional end-to-end congestion control such as that carried out by TCP becomes unsuitable as it is unable to take advantage of such in-network functionalities. We design a congestion control policy that uses explicit congestion notifications from network routers and rate control at traffic sources. The hop-by-hop reliability provided in MF simplifies end-to-end reliable delivery of wireless/mobile data, but often requires routers to keep per-flow queues to carry out congestion control which could become impractical in the presence of a large number of flows. Our approach builds on a per-interface queueing scheme, and we show through simulation that it is able to substantially improve delay, fairness, and scalability with only ≤6% link utilization degradation, compared with a per-flow queueing based scheme.