Impacts of application lookahead on distributed network emulation

Abstract

Large-scale and high-fidelity testbeds play critical roles in analyzing large-scale networks such as data centers, cellular networks, and smart grid control networks. Our prior work combines parallel simulation and virtual-time-integrated emulation, such that it offers both functional and temporal fidelity to the critical software execution in large scale network settings. To achieve better scalability, we have developed a distributed emulation system. However, as the number of computing servers grows, so does too the synchronization overhead. Application lookahead, the ability to predict future behaviors of software, may help reducing overhead for performance gain. In this paper, we study the impacts of application lookahead on our distributed emulation testbed. We find that application lookahead can greatly reduce synchronization overhead and improve speed by up to 3 times in our system, but incorrect lookahead may affect application fidelity to different degrees, depending on application sensitivity to timing.