Scenario-Specific Topology Reduction in Network Simulations

Our current ability to perform packet-level simulations of large-scale networks is limited by factors such as processing speed and memory resources. Many approaches have been proposed to reduce the complexity of simulation models in a manner that reduces the computing resources required while preserving the fidelity of the simulation. In this paper we consider topology reduction as a means to reduce the resource requirements of network simulations. We show that it is possible to reduce topologies of simulation models while preserving certain classes of metrics. We do this in the particularly challenging environment where closed loop traffic (e.g., TCP) is being simulated in the network. The reductions are scenario specific in the sense that the desired performance metrics of a network simulation experiment dictate the feasibility of particular reduction methods. We use a set of experiments representing simulation scenarios to demonstrate our approach.