Internet behavior is becoming more complex due to ever-changing networking technologies and applications. Thus, understanding and controlling the complex behavior of the Internet are important for designing future networks. One of the complex behaviors of the Internet is traffic dynamics. Previous studies revealed that flow control in the transport layer affects the traffic dynamics of the Internet. However, it is not clear how the topological structure impacts traffic dynamics. In this paper, we investigate packet delay dynamics and traffic fluctuation in ISP router-level topologies where the degree distribution exhibits a power-law nature, and the nodes interact via end-to-end feedback control functionality. We show the packet delay dynamics of the BA topologies generated by the Barabási-Albert (BA) model and the ISP router-level topologies. Simulation results show that the end-to-end delay distributions exhibit a heavy tail in the TCP model. Moreover, the number of links with highly fluctuating queue length increases dramatically compared to that in the stop-and-wait model. Even in this case, the high-modularity structures of the ISP topologies reduce the number of highly fluctuating links compared with the BA topologies.