Management Framework for the Internet of Consumer Electronics

The rapid growth of Internet of Consumer Electronics (ICE) devices has exposed cloud computing’s limitations in meeting emerging application demands such as low latency, high reliability, privacy, and efficient resource use. Fog computing offers a promising alternative; however, no comprehensive study has addressed predicting the MIPS (Millions of Instructions per Second) performance of heterogeneous nodes for scheduling. This paper proposes a framework for predicting fog node CPU performance based on calibration regression. Benchmark tests are conducted pre-deployment on ICE devices under controlled CPU and memory loads. Unlike historical data approaches, this method enhances the accuracy and efficiency of machine learning-based MIPS prediction. A three-module framework for node calibration, resource collection, and performance prediction is introduced. Using a custom load generator, we collect data on CPU usage, memory usage, temperature, and execution time to train four regression models. The model with the best trade-off in accuracy and speed is selected. Results show CPU usage is strongly negatively correlated with MIPS, while memory usage is less significant. The Least Absolute Deviation (LAD) model achieves 97.24% prediction accuracy with an average prediction time of 7.8 ms, confirming the effectiveness and efficiency of the proposed approach.