Towards more realistic network simulations: Leveraging the system-call barrier

Abstract

Network simulations play a substantial role in evaluating network protocols. Simulations facilitate large-scale network topologies and experiment reproducibility by bridging the gap between analytical evaluation and real-world measurements. A recent trend in discrete event network simulations is to enhance simulation realism and reduce duplicate implementation efforts by maximizing code reuse. Despite such efforts, it is not yet possible to run arbitrary network applications in state-of-the-art network simulators. As a consequence, researchers are required to maintain separate protocol implementations: one for real-world measurements and one for simulations. We review existing approaches that maximize code reuse in simulations, compare their limitations, and propose a novel architecture for protocol simulation that overcomes those restrictions.