Router-Assisted Congestion Control for Wired and Wireless High-Speed Networks.

摘要

The resource management mechanisms such as congestion control protocols and flow rate adaptation algorithms are critically important to make the Internet work efficiently and stably. As more and more high-speed links, lossy links, long-delay links and variable-delay links are widely deployed in the Internet, traditional end-to-end congestion control schemes exhibit several shortcomings, such as poor utilization of high-speed links, unfair bandwidth allocation among flows with different round-trip times (RTTs), incorrect interpretation of bit-error packet loss as congestion, and slow responsiveness to fast-changing physical conditions of mobile nodes. To address these problems, we develop flexible and adaptive cross-layer schemes that optimize the performance (throughput, latency, fairness, etc.) of the whole networking system. We propose Quick Flow Control Protocol (QFCP) as a router-assisted congestion control protocol for high bandwidth-delay product networks. It allows flows to start with high initial sending rates indicated by routers along the path and to converge to the fair-share sending rate quickly based on feedbacks from routers. We next extend QFCP to wireless networks so that it can distinguish bit-error loss from congestion loss and can probe the unknown bandwidth capacity of wireless links to calculate the router feedback. Time-constraint flows are very common in vehicular communications. They have fixed start and stop times and try to maximize the transferred data volume during the limited connection time. We find that the traditional fairness concept solely based on instantaneous flow rates is not suitable for this scenario. Therefore, we propose new practical bandwidth sharing scheme s for transferring data with fast-moving wireless nodes such as vehicles based on the utility fairness. We also validate and evaluate our developed schemes under various environments and conditions comparing with other existing protocols. Performance evaluation is presented here and the results of our protocols are promising.