Internet congestion control with network measurement support

The Transmission Control Protocol (TCP) manages a large fraction of Internet traffic. It relies on loss or delay measures to infer the existence of congestion (or lack thereof), and reacts by increasing/decreasing a flow transmission rate. Without an accurate estimate of the available bandwidth on the communication channel, TCP is either too aggressive leading to more congestion, or too conservative leading to low channel utilization. Recent advances in bandwidth estimation tools can help TCP achieve better channel efficiency while maintaining fairness among competing flows. In this paper, we propose a rate-based congestion control algorithm that relies on a bandwidth estimation tool, Forecaster, and investigate its performance. Experimental results reveal its superiority to existing algorithms in efficiently using the available bandwidth and in its resilience to lossy channels, making the algorithm suitable for common HTTP short flows and for communication over network paths with wireless hops.