Drift: A highly condensed emulation framework for mobile nodes in server clusters

Xinyu Zhang, Baochun Li

Abstract: Prototyping application-layer algorithms in wireless networks is a lengthy and challenging process, involving either programming within a specific simulation platform, or deploying a real testbed that is neither flexible nor scalable. In this paper, we present Drift, a high-performance wireless emulation testbed that takes advantage of the benefits of both simulation and real implementation, while trying to avoid their drawbacks. As a highly condensed emulation infrastructure, Drift makes it possible to rapidly develop and validate large-scale wireless network application-layer protocols within a cluster computing environment. Unlike existing emulation testbeds, Drift features a fully decentralized architecture, an efficient message processing unit, and more accurate network models for mobile wireless nodes. It balances the fundamental trade-off between scalability and emulation accuracy, focusing on maximizing scalability with minimal loss of accuracy. Through baseline comparison and extensive experiments, we find that Drift is able to accommodate thousands of emulated nodes per server host (in contrast to only tens of nodes typically seen in existing emulation tools), while maintaining comparable accuracy to packet level simulators. Drift will be released as an open source platform.