The Influence of Realism on Congestion in Network Simulations
Kevin L. Mills, Christopher E. Dabrowski
Many researchers have used simulation to investigate the spread of congestion in networks. These researchers often find that congestion can be modeled as a percolation process, spreading slowly under increasing load until a critical point. After the critical point, congestion spreads quickly throughout the entire network. The researchers also identify various measureable signals that arise around the critical point. These findings appear quite promising as a theoretical basis for monitoring regimes that network operators could deploy to warn of impending congestion collapse. Yet questions surround the extant research because the findings arise from models that are quite abstract. Such models bear little resemblance to networks deployed based on modern technology. We explore these questions by examining the influence of realism on the spread of congestion in network simulations. We begin with an abstract network simulation, taken from the literature, and add elements of realism in various combinations, culminating with a high-fidelity simulation, also taken from the literature. By comparing patterns of congestion among combinations, we make four main contributions. First, we illustrate that congestion spread in abstract network models differs significantly from spread in realistic models. Second, we show that models investigating network congestion must include specific elements of realism before acceptable engineering findings can be established. Third, we identify the influence of specific elements of realism on congestion in network simulations. Finally, we demonstrate an effective means to compare congestion patterns among network simulations comprising diverse configurations. We hope our contributions lead to better understanding of the influence of realism on congestion in network simulations, and to improved dialog throughout the diverse community of researchers who rely on network simulations.