TCP-SDF: Transport control to achieve smooth data flow at a stable throughput

Various large-scale network applications demand different levels of stable bandwidth to perform bulk data movement, which is not adequately supported by current transport methods. We propose a novel end-to-end transport control protocol, (TCP-SDF), which achieves smooth data flow at a stable good put level. TCP-SDF dynamically adjusts the congestion window based on the estimate of the current sending rate, good put, and target level. The adjustment terms are weighted by the step-sizes, which are non-negative numbers that eventually die down to zero as specified by the classical Robbins-Monro Stochastic Approximation conditions. TCP-SDF decreases the congestion control window size by a factor that is proportional to the estimated number of packets in the bottleneck queue in the presence of network congestion. TCP-SDF can be readily implemented in practice as a loadable kernel module in Linux. We demonstrate the superior performance of TCP SDF under a range of scenarios using experimental measurements.