Reviewer: 1
Comments to the Author
Ths paper discusses a new method for delay-based AIMD congestion control. Although  authors propose a simple mechanism, this variant of Additive
Increase/Multiplicative-Decrease (AIMD) algorithm exibits very good co-existence of the delay-based and loss-based flows as measured by average throughput.

Authors provide analytical and simulaton results to validate presented alghorithm.
From simulation and experimental results it is obvious that this simple mechanism allows delay- and loss-based (AIMD) TCP flows to compete fairly with each other.

This paper could be very important for developmento of delay-based congestion control algorithms.


Reviewer: 2
Comments to the Author
The authors propose a simple mechanism that allows delay-based and loss-based (AIMD) TCP flows to compete fairly with each other. The paper, however, leaves the following points that should be addressed in order to be publishable.

1. The paper does not provide sufficient rigorous analysis to show the stability and performance of the proposed algorithm. In particular, the authors do not present a clear proof for Theorem 1 that is the main result of the paper. The authors should have allowed more space for the rigorous analysis instead of the simulation results.

2. There is no comparison of the proposed algorithm with other related algorithms in the simulation results. Besides, the authors do not present any criterion to judge the superiority of the proposed algorithm in the discussion of the simulation results. Therefore, it is quite unreasonable to conclude that the proposed algorithm shows a good co-existence property without any comparison or acceptable criterion.


Reviewer: 3
Comments to the Author
I found the potential use of a different response function (in Fig. 1) to delay to be an interesting and useful idea.

However, the authors haven't provided enough details both in theory and simulations to make the paper convincing. The authors should provide proof of theorem 1 and provide more data from simulations.

Some specific questions: How do you choose Wo? What is the impact on choosing a larger or smaller value for this parameter? Does this depend on the number of flows, link bandwidth etc.

The authors should provide some simulations with flows starting at different times --this is one of the difficulties in using the delay-based mechanisms -how do all the flows converge on a shared view of the bottleneck link -this is necessary for ensuring fairness.

The authors may want to comment more on the approach taken here and the approach taken in [11] --both seem to be targeting the same problem --coexistence of delay-based and loss-based protocols.