Dynamics of Feedback-Induced Packet Delay in Power-Law Networks

Abstract

Measurement studies of the Internet topology revealed that the degree distributions of the Internet topologies follow a power law. That is, existing probability of nodes having k links is proportional to k^{−γ}. However, only the power-law degree distribution does not determine network-level performance of networking methods. That is, the structural characteristics of topologies other than the degree distribution are essential to discuss the behavior of networking methods. In this paper, we investigate traffic dynamics on ISP’s router-level topology where the degree distribution exhibits power-law attribute and each of the nodes interacts via end-to-end flow control functionality. We show packet delay dynamics on the BA topology generated by the BA model and the ISP’s router-level topology. Simulation results show that the end-to-end packet delay distributions exhibit a heavy tail by the TCP model. Moreover, the number of links that are highly fluctuated is more than twice comparing to the results of the stop-and-wait model. Even in this case, the modularity structure of the ISP topology reduces the number of highly fluctuated link comparing to the results of the BA topology.