Fluid-Based Analysis for Understanding TCP Performance on Scale-Free Structure

Abstract

Scale-free structure is one of the most notable properties of the Internet as a complex network. Many researchers have investigated the end-to-end performance (e.g., throughput, packet loss probability, and round-trip time between source/destination nodes) of TCP congestion control mechanisms, but the impact of the scale-free structure on the TCP performance has not been fully understood. In this paper, we analyze the TCP performance on a scale-free tree whose strength of the scale-free property can be adjusted by a parameter. A scale-free tree represents the communication kernel for investigating a scale-free network since TCP mainly transmits packets on a shortest path between TCP source/destination nodes, and most shortest paths are included in the scale-free tree. Our numerical results show that the scale-free structure of a network improves the TCP performance, and that such performance improvement is caused by a reduction in the average path length and also a reduction of the traffic intensity at the bottleneck link. Furthermore, we confirm the validity of our analysis through a comparison with an optimization-based analysis.