An Experimental Study on Congestion Control in Wireless and Wired Networks

This paper presents an extensive experimental analysis of two TCP-friendly congestion controls recently proposed in the literature, placing a special emphasis on the wireless setting. This work is motivated in part by the consideration that, to our knowledge, the majority of the results in this area are validated by simulation analysis rather than by experimentation. Testing these algorithms in real environments can help to verify their actual effectiveness for possible employment over the Internet. To reach our goal, we have implemented one representative algorithm among the so-called window-based TCP-friendly congestion controls, namely, the general additive increase multiplicative decrease (GAIMD) strategy, and used the publicly available code for the most popular control of the rate-based family, i.e., the TCP-friendly rate control (TFRC). We have then evaluated their TCP-fairness and smoothness in a test-bed featuring Cisco WAN equipment, as well as in an IEEE 802.11g wireless local area network (WLAN). Based on our results, we first propose and justify the tuning of TFRC to avoid a weird behavior that in the wireless environment this control occasionally exhibited. We next show that the GAIMD strategy reveals non-negligible scalability and smoothness problems that limit its performance, mainly in the radio setting. We empirically demonstrate that its increase/decrease rules, based on a TCP-Reno analytical model, do not guarantee an adequate performance when GAIMD competes with TCP-Sack, a de facto standard for current TCP implementations.