F-TCP

Abstract

We describe F-TCP, a new TCP for file transfer in high bandwidth-delay product networks. TCP is not suitable for file transfer application in high bandwidth-delay product networks. That is, a protocol should not take away too much bandwidth from standard TCP flows while utilizing the full bandwidth of high-speed networks. This paper presents another important constraint, namely small buffer problems. Small socket buffers at the end-hosts limit the effective window of the transfer, and thus the maximum throughput. Existing schemes have a severe small buffer problems because high bandwidth-delay product networks need a large socket buffer. The proposed mechanism, called F-TCP, is based on direct kernel-to-kernel file transfer architecture. The key idea is that no need of in-order delivery in file transfer process. F-TCP needs only small buffer by allowing out-of-order file transfer with asynchronous loss recovery. Experimental results in several high bandwidth-delay product paths show that F-TCP provides consistently a significant throughput increase (18% to 80%) compared to previous TCP approach. We expect that F-TCP will be mostly useful for application such as FTP and P2P in high-speed wide-area networks.