查看文章

This paper presents a new congestion control algorithm of TCP, called TCP-Fusion, and provides its extensive evaluation results through simulations and implementations. Recently, towards high-speed networks with large bandwidth delay product, a number of different approaches have been proposed to improve TCP performance. However, their potential unfriendliness to TCP-Reno encumbers their wide deployment in the Internet because TCP-Reno is already widely deployed. Most recently, to satisfy efficiency and friendliness tradeoffs of TCP, new approaches combining a loss-based protocol and a delay-based protocol have been proposed, such as TCP-Adaptive Reno and Compound TCP. Our TCP-Fusion also belongs to this category and tries to utilize the residual capacity effectively without impacts on coexisting flows, ie TCP-Reno flows. To achieve this purpose, TCP-Fusion exploits three useful characteristics of TCP-Reno, TCP-Vegas and TCP-Westwood in its congestion avoidance strategy. In short, congestion window of TCP-Fusion is decreased without causing too drastic reduction and is increased with smart adaptability to coexisting TCP-Reno flows according to the congestion level measurement estimated from RTT. Our implementation and simulation results show that TCP-Fusion can obtain the highest throughput among existing TCP variants when there is unused residual capacity while its friendliness to the TCP-Reno is sufficiently satisfied, otherwise, it shares the same bandwidth to coexisting flows.