Fairness Measurement Procedure for the Evaluation of Congestion Control Algorithms.

Congestion is one of the most critical issues im-pacting the performance of Internet networks, hence the need of Congestion Control Algorithms (CCAs) to either prevent or remove it. Nevertheless, CCAs may affect the network fairness given that the transport behavior can drastically change in function of the CCA (e.g., Performance-oriented Congestion Control (PCC), Bottleneck Bandwidth and Round-trip propagation time (BBR), Reno) used in the endpoints. New transport protocols such as Quick UDP Internet Connections (QUIC) allow to easily customize their CCA, which increases the types of CCAs on the Internet. Evaluating the fairness among competing CCAs is thus essential. In this work, we focus on determining an impartial Fairness Measurement Procedure (FMP) capable of assessing equity when several CCAs compete in a bottleneck during a given period of time. In order to validate the proposed mechanism, we perform an exhaustive evaluation of various competing CCAs within multiple network configurations while varying latency, packet loss, queuing policy and buffer size. Results show that the grade of fairness achieved by two different CCAs depends on the network configuration. This claim confirms the relevance of the proposed fairness measurement procedure.