Decomposing RTT-unfairness in transport protocols

In this paper, we consider RTT-unfairness in most existing end-to-end congestion-control protocols, in which transfers with smaller RTTs are allocated a higher share of the bottleneck bandwidth. We consider the congestion-control mechanisms used by TCP NewReno and identify three aspects that introduce an RTT-based bias in different ways. Past attempts at alleviating RTT-unfairness mostly consider only one of these aspects. We use a first-principles approach for deriving a set of simple scaling factors that can be used to remove the RTT-bias. We conduct empirical evaluations to study how fair in practice would be a hypothetical protocol that employs all of these scaling factors. We discuss the practicality of implementing such a protocol.