Automated power modeling of computing devices: Implementation and use case for Raspberry Pis

Abstract

Monitoring the power consumption of smart and connected devices is a challenging task with heterogeneous devices and a variety of hardware and software configurations. In order to accurately monitor these devices and follow the speedy changes in their configuration, a new approach is needed. In this paper, we present an automated architecture and approach to empirically generate power models for a large set of devices. Our approach allows conducting automated benchmarks to collect power data and metrics, generating or updating accurate power models, and allowing software tools to query and retrieve the most accurate and up-to-date power model of a specific device configuration. We also present a proof-of-concept implementation for modeling the power consumption of Raspberry Pi devices. Finally, we conduct a comprehensive experiment, modeling the entire current lineup of Raspberry Pi devices with error margins as low as 0.3%, and then we discuss the impact of multiple device configurations on power consumption.