Large-scale and high-fidelity wireless network simulation and emulation

Abstract

Large-scale networks, such as the Internet, cellular networks, play critical roles in today's life. The advancement of large-scale computer and communication networks heavily depends on the successful transformation from in-house research efforts to real productions. When studying these network systems, modeling is an important and useful approach as it allows studies not physically realizable. In the domain of wireless networks, simulations are usually used to study new or existing designs, as it is economically and technically expensive to implement those designs using real hardware. However, evaluation based on any methodologies other than actual measurements on actual networks raises questions of fidelity, due to necessary simplifications and assumptions. In this dissertation, we present a system for large-scale and high-fidelity wireless network simulation and emulation. Our system provides high-fidelity in three areas. 1) High physical layer fidelity: we use sophisticated radio propagation models such as ray-tracing, validated by an anechoic chamber. 2) High application functional fidelity: we allow unmodified and compiled application code to run inside our system, minimizing the modeling error of software behavior. 3) High application temporal fidelity: we provide virtual timestamps to the virtual machines rather than using the wallclock time, making them perceive time as if they were running concurrently in real world. Besides high-fidelity, our system can handle large-scale network scenarios in reasonable speed, by making the simulation and emulation run in parallel across multiple machines. Application lookahead, the ability to predict future behavior of software, can help further improve speed and scalability by reducing the frequency of synchronization. However, it may affect fidelity as lookahead could be wrong due to software complexity and runtime uncertainty. We extensively study the impacts of lookahead on our system to provide guidelines of when lookahead may be used.