Characterization of complex behaviors of TCP/RED computer networks based on nonlinear time series analysis methods

Packet-level observations are representative of the high sensitivity of TCP/RED computer network behavior with respect to network/RED parameter variations. That is, while we do not have any control on network parameters, mis-choosing of the RED parameters results in complex non-periodic oscillations in the router queue length that may damage the Quality of Service requirements. Characterizing the nature of such behaviors, however, helps the network designers to modify the RED design method in order to achieve better overall performance. In this paper, we first investigate the effect of variations in different RED parameters on the network behavior and then seek for the origin of such complex behaviors. For this purpose, different linear and nonlinear time series analysis methods have been applied to long-duration ns-2 packet-level data traces of a homogeneous TCP/RED network. The results of the analysis confirm that the complex behavior of the network can be represented by a nonlinear stochastic second-order process and it is not due to a deterministic chaos or a non-stationarity in the network.