Rebuttal to “Experimental Evaluation of Cubic- TCP” by Leith, Shorten and McCullagh

We are writing this paper as a response to a paper appeared in PFLDnet 2007 authored by Leith et al. on the analysis of CUBIC [1] (hereafter called LSM). There are a few debatable issues that might mislead readers about the behavior of CUBIC. In general, the authors presented their observations on the performance of CUBIC under a special set of testing environments. However, as this paper is about CUBIC, a more balanced treatment on whether these observations are applicable uniquely to CUBIC would clarify potential misconceptions about CUBIC that LSM might cause – in other words, the authors did not discuss whether the observations are really indications of unique features of the protocol or the “normal” behaviors commonly observed in standard TCP and other HS protocols under the same environments. Note that there exists no absolute performance metrics that can define a protocol to be “unfair” or “slow” as these are all relative terms. Thus, their discussions alluding to the “gross unfairness” and “very slow convergence” of CUBIC without proper comparison with other protocols might lead to a misconception that CUBIC is really an incorrectly behaving protocol. Although the authors choose to omit such a balanced discussion or simply did not experiment with other protocols for comparison, a proper qualification of their statement, e.g., “so and so behaviors MAY NOT be unique to this protocol,” would prevent such a misconception. In addition, the authors seem to use CWND extensively as a metric to measure unfairness. But this metric can be misleading as CWND is not the same as transmission rate especially under large buffers. The purpose of this rebuttal is to point out the areas of the document that might mislead readers. As a disclaimer, this rebuttal is not reviewed by any third-party.