Slowdown as a Metric for Congestion Control Fairness

Abstract

The conventional definition of fairness in congestion control is flow rate fairness. However, Internet users typically care about flow completion times (FCTs) and flow rate fairness does not lead to equitable FCTs for different users. Therefore, we reconsider what it means for congestion control to be fair and posit a novel stance on fairness: it is fair when no flow unnecessarily prolongs another flow. Based on this stance, we propose an evaluation framework for congestion control fairness that uses slowdown (normalized FCT) as the metric.
We demonstrate the usefulness of our framework through surprising experiment results: in theory, prioritizing short flows should outperform fair queueing, but we show that this is not the case due to slow start dominating short flows. The framework can also analyze traditional flow rate fairness; we do so and verify well-known "fairness" issues, but additionally, we show that flow rate unfairness does not induce slowdown and is thus not a problem per se.