TCP-Adaptive in High Speed Long Distance Networks

With the development of high performance computing and increasing of network bandwidth, more and more applications require fast data transfer over high-speed long-distance networks. Research shows that the standard TCP Reno cannot fulfill the requirement of fast transfer of massive data due to its conservative congestion control mechanism. Some works have been proposed to improve the TCP throughput performance using more aggressive window increasing tactics and obtain substantial achievements. However, they cannot be strictly proved to be comprehensively suitable for high-speed complex network environments. In this paper, we propose TCP-Adaptive, an adaptive congestion control algorithm adjusting the increasing congestion window dynamically. The algorithm improves logarithmic detection procedure for available bandwidth in the flow path by distinguishing the first detection in congestion avoidance and retransmission timeout. On the other hand, an adaptive control algorithm is proposed to achieve better performance in high-speed longdistance networks. The algorithm uses round trip time (RTT) variations to predict the congestion trends to update the increments of congestion window. Simulations verify the property of TCP-Adaptive and show satisfying performance in throughput, RTT fairness aspects over high-speed longdistance networks. Especially in sporadic loss environment, TCP-Adaptive shows a significant adaptability with the variations of link quality.