Data-driven Software-based Power Estimation for Embedded Devices

2024-07-03

IEEE Transactions on Sustainable Computing

AI Summary

To address the widespread lack of hardware-level power measurement capability in IoT embedded devices, this paper proposes a lightweight software-based power estimation algorithm that integrates an external low-cost USB power meter with data-driven modeling. Our method innovatively combines long-term physical calibration, fine-grained runtime resource feature extraction (e.g., CPU/GPU utilization and frequency), and a lightweight regression model to enable real-time, program-level instantaneous power prediction directly on-device. A custom Linux kernel module ensures synchronized USB sampling and feature acquisition. Evaluated on Jetson Nano and Raspberry Pi platforms, the approach achieves an average estimation accuracy of 92%, with minimal deployment overhead and over 90% reduction in total cost compared to hardware-integrated solutions. Crucially, it requires no hardware modification, offering a high-accuracy, low-cost, and easily deployable power profiling paradigm for resource-constrained IoT devices.

Technology Category

• Machine Learning: Hardware-aware ML
• Application Domains: Internet of Things, Sensor Networks & Smart Cities
• Intelligent Robots: State Estimation

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Security and Privacy: Large-scale security measurements

Abstract

Energy measurement of computer devices, which are widely used in the Internet of Things (IoT), is an important yet challenging task. Most of these IoT devices lack ready-to-use hardware or software for power measurement. In this paper, we propose an easy-to-use approach to derive a software-based energy estimation model with external low-end power meters based on data-driven analysis. Our solution is demonstrated with a Jetson Nano board and Ruideng UM25C USB power meter. Various machine learning methods combined with our smart data collection&profiling method and physical measurement are explored. Periodic Long-duration measurements are utilized in the experiments to derive and validate power models, allowing more accurate power readings from the low-end power meter. Benchmarks were used to evaluate the derived software-power model for the Jetson Nano board and Raspberry Pi. The results show that 92% accuracy can be achieved by the software-based power estimation compared to measurement. A kernel module that can collect running traces of utilization and frequencies needed is developed, together with the power model derived, for power prediction for programs running in a real environment. Our cost-effective method facilitates accurate instantaneous power estimation, which low-end power meters cannot directly provide.

Problem

• Estimating power consumption in IoT devices without built-in measurement tools
• Developing software-based energy models using low-end power meters and machine learning
• Achieving accurate real-time power prediction for embedded systems like Jetson Nano

Innovation

• Data-driven software-based energy estimation model
• Machine learning with smart data collection
• Kernel module for real-time power prediction