Studying Congestion Control with Explicit Router Feedback Using Hardware-based Network Emulator

Congestion control with explicit router feedback, such as TCP Quick-Start and XCP, is a promising way to enhance the performance of data transport especially in high-speed networks. When designing such transport protocols, the tradeoff between the granularity of feedback information and additional cost of routers for feedback processing and state keeping should be discussed. However, the feasibility is never convinced in terms of its implementation and deployability because there is no experimental high-speed network environment to evaluate the additional cost. In this paper, we study congestion control with explicit router feedback using hardware-based high-performance network emulator. The network emulator consists of a general-purpose network processor and two Gigabit Ethernet (GbE) ports, where new explicit router feedback functions can be easily implemented and evaluated. The goal of this paper is to gain an insight into feasibility of explicit router feedback and to give a possible direction to design such congestion control mechanisms. To this end, we first discuss the deployability of fine-grained explicit router feedback through the experience of developing the high-performance network emulator and evaluations of the performance. In addition, based on the discussion, we propose a unified explicit router feedback framework. The framework can be applied to a broad class of transport protocols and can provide fine-grained feedback information fundamentally required by end-hosts to design a novel congestion control mechanism or to significantly enhance the performance of traditional congestion control in specific situations. As one possible application of the unified explicit router feedback framework, we show the extension of TCP Limited Slow-Start [2]. Through the implementation and evaluation, we make sure that a sender can shift from slow start phase to congestion avoidance phase without any packet drop, and can achieve more effective bandwidth utilization.