WattEdge: a holistic approach for empirical energy measurements in edge computing

Of the main challenges to keep the edge computing dream alive is to efficiently manage the energy consumption of highly resource-limited nodes. Past studies have limited or often simplistic focus on energy consumption factors considering computation or communication-only solutions, questioned by either costly hardware instrumentation or inaccurate software-specific limitations. With this gap in mind and the wide adoption of single-board computers (SBCs) such as Raspberry Pis in edge, in this paper, we propose a novel holistic and accurate energy measurement approach in edge computing. Exploring a Test and Learn strategy, (1) we firstly perform a comprehensive analysis of identifying factors affecting energy consumption of edge nodes; (2) we develop and utilize WattEdge, a standard framework to evaluate the identified factors; (3) we conduct extensive empirical experiments on Raspberry Pis to thoroughly and uniformly assess the significance of each factor, thereby proposing an all-inclusive energy model. Wattedge is able to measure energy consumption factors such as CPU, memory, storage, a combination of them, connectivity, bandwidth usage, and communication protocols, as well as energy sources such as batteries. The results specifically warn us of the necessity of considering previously underestimated factors such as connectivity. A Smart Agriculture use case is implemented to validate the performance of the energy model, demonstrating a 95% accuracy.