Network Device Telemetry Modeling for Real-Time Power Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for monitoring network device power consumption are costly, bulky, and impractical for widespread use, leading to inaccuracies and design constraints, especially in smaller devices like access points and cameras, and external power monitors are large, pricey, and lack real-time data capabilities.
Innovation Solution
A telemetry-based approach using a machine learning model trained on device telemetry data, including motherboard temperature, CPU temperature, and memory capacity, to predict power consumption without adding external sensors, enabling cost-effective and accurate power consumption prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If external power sensors are used to monitor network device power consumption, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent creates a virtual copy of power consumption data by training a machine learning model to predict power consumption values based on telemetry parameters. Instead of using physical sensors to directly measure power consumption, the system copies the information from readily available telemetry data (CPU usage, memory usage, temperature, etc.) through a trained predictive model, thereby achieving accurate power consumption measurement without adding physical sensing components to the network device.
2Loss of information
If external power monitors are used, then power consumption data is obtained, but the monitors are large, pricey, and lack real-time data capabilities
Solution Approach 1:
The patent replaces the mechanical/physical power monitoring system with a software-based machine learning approach. Instead of using physical power monitors that require hardware integration, the system substitutes a trained machine learning model that processes existing telemetry data to generate real-time power consumption predictions. This substitution eliminates the need for bulky, expensive external monitoring hardware while providing continuous real-time data capabilities.
3Measurement precision
If conventional power monitoring methods are used, then power consumption is measured, but costs increase and design constraints are imposed on smaller devices
Solution Approach 1:
The patent employs a software-based solution that leverages existing, inexpensive telemetry infrastructure already present in network devices. Rather than requiring costly, custom-designed hardware components for power monitoring, the system uses readily available telemetry data processed through a machine learning model. This approach dramatically reduces manufacturing costs and design complexity, making power consumption monitoring feasible for smaller, cost-sensitive devices like access points and cameras.
Data Source
AI summary
Devices, systems, methods, and processes for telemetry-based device power consumption prediction are described herein. Values of power consumption and telemetry parameters associated with a network device are collected over a time period. Using at least one telemetry parameter, various engineered parameters are generated. From all the collected telemetry parameters and the engineered parameters, a set of model parameters is selected for model development. A machine learning (“ML”) model is then trained to determine a correlation between the values of the set of model parameters and the power consumption of the network device. When the network device is in the field, device telemetry data is sensed. Based on the device telemetry data, values corresponding to the set of model parameters are determined and provided as input to a trained ML model. Device power consumption is predicted based on an output of the trained ML model for the input values.