Fluid-flow approximation is an approach to modeling and evaluating the performance of vast computer networks. Due to varying traffic and performance of transmission protocols reacting to traffic overloads, computer networks are in a permanent transient state. The fluidflow method’s main advantage is its ability to analyse these transient states. The article reviews and organises several versions of this approach, indicating a few errors. The main reason for these errors is confusion or lack of distinction between the two versions of the Internet Protocol - when the queue of packets at a node is too long, they may be destroyed or only marked as redundant. The paper compares and evaluates these fluid-flow approximation models with mild and aggressive settings of RED parameters. The authors build a software system with hitherto unprecedented capabilities regarding the size of the networks to be analysed and with innovative way of organising the calculations. The paper shows how large differences imprecise assumptions can introduce in quantitative results.

Establishing the proper values of controller parameters is the most important thing to design in active queue management (AQM) for achieving excellent performance in handling network congestion. For example, the first well known AQM, the random early detection (RED) method, has a lack of proper parameter values to perform under most the network conditions. This paper applies a Nelder-Mead simplex method based on the integral of time-weighted absolute error (ITAE) for a proportional integral (PI) controller using active queue management (AQM). A TCP flow and PI AQM system were analyzed with a control theory approach. A numerical optimization algorithm based on the ITAE index was run with Matlab/Simulink tools to find the controller parameters with PI tuned by Hollot (PI) as initial parameter input. Compared with PI and PI tuned by Ustebay (PIU) via experimental simulation in Network Simulator Version 2 (NS2) in five scenario network conditions, our proposed method was more robust. It provided stable performance to handle congestion in a dynamic network.