An important factor in the robustness of the Internet is end-to-end transmission control protocol (TCP). Two well-known congestion avoidance mechanisms are TCP Reno and TCP Vegas. Hosts using TCP Reno may not react in time to the congestion. TCP Vegas compares the actual transmission rate with the expected rate, and adjusts window size accordingly. Flows with distinct start times may get different shares of bandwidth. Thus, TCP Vegas suffers from the fairness problem.

The authors propose an enhanced TCP Vegas scheme (EVA), with three revisions to improve its performance. These revisions include replacing BaseRTT with round trip time (RTT), to improve the competitiveness of newcomer flows; detecting how fast acknowledgments return; and the acceleration of return speed. TCP EVA improves TCP Vegas in two aspects: fairness and quick adaptation to the network condition.

The impacts of the three revisions on the fairness improvement of TCP Vegas are verified by factorial designs. Simulation results show an improvement in fairness when using TCP EVA. Each of the three factors has an impact on the results. The authors show that EVA achieves better fairness under various network conditions, such as congestion in the reverse path, on-off traffic, and flows with different RTTs. However, the behavior of EVA under multiple disturbances seems to be chaotic compared to TCP Vegas, and no explanation about this was given by the authors.

In summary, the contributions of this paper are well stated and well supported by experimental results. The paper is also well written, and properly structured.