An Application Based Rate Controllable TCP for High Performance Bulk Data Transfer Guodong Wang Computer Network Information Center, Chinese Academy of Sciences Graduate University of Chinese Academy of Sciences Beijing, China wangguodong@cstnet .cn Yongmao Ren Computer Network Information Center, Chinese Academy of Sciences Beijing, China renyongmao@cstnet.c n Xiaodan Zhang Computer Network Information Center, Chinese Academy of Sciences Graduate University of Chinese Academy of Sciences Beijing, China zhangxiaodan@cstnet .cn Jun Lee Computer Network Information Center, Chinese Academy of Sciences Beijing, China jlee@cstnet.cn ABSTRACT This paper designs a novel congestion control algorithm named application based rate controllable TCP (ABC-TCP), whose transfer rate can be adjusted by the applications. If the required transfer rate doesn’t meet the requirement of the application, the sender’s congestion control window (CWND) will increase aggressively and if it gets the required transfer rate the congestion control window will stop growing, in which way to ensure the required transfer rate. ABC-TCP is implemented by modifying TCP code in Linux. Experiments on an FLDnet test-bed show that ABC-TCP is able to keep a stable throughput and satisfy critical application’s needs. Keywords High Speed Network; Transport Protocol; Congestion Control; Bulk Data Transfer. 1. INTRODUCTION Moving bulk data quickly over high-speed data networks is a requirement for many sciences research such as the high energy physics, molecular biology, e-VLBI [1] in astronomical observation fields and others [2]. However, data transfers with such end-to-end performance are still not being widely realized over wide-area networks, despite the availability of wide-area connections. Traditional TCP protocols used in these networks are not suitable in meeting the high-throughput and large RTT. Several promising new protocols have been put forward: HSTCP [3], STCP [4], BIC [5], and CUBIC [6]. These protocols modify the increase and decrease rule of Reno congestion control to be more aggressive when they work in the high BDP networks. In the study and use of these protocols we found a problem that all of these TCP variants do not stop the increase of CWND until the network congested and then the CWND dropped. After that the TCP have to spend a long time to increase the CWND to the needed level. In the process of fluctuation of the CWND from up and down, lots of valuable bandwidth is wasted, which is a main reason for the poor performance of the large BDP network [7]. Fairness of TCP is a main focus in the previous promoted TCP variants [3, 6]. [3] gives us a common idea that under high loss rate regions TCP variants should behave like TCP. The use of a new TCP variant should not unfairly affect the most common network flows (namely TCP). Fairness is important to keep all flows in order, but it is not an absolute rule in transmitting emergency files. In some scenes