NetAP-ML: Machine Learning-Assisted Adaptive Polling Technique for Virtualized IoT Devices.

Area of Science:

• Computer Science
• Machine Learning
• Edge Computing

Background:

• Optimizing Internet of Things (IoT) device performance in edge computing requires efficient polling interval selection.
• Current methods often involve extensive search, leading to performance degradation.

Purpose of the Study:

• To introduce NetAP-ML, an adaptive polling technique using machine learning to optimize polling intervals for IoT edge devices.
• To reduce the search space and minimize performance loss during optimization.

Main Methods:

• Utilized a machine learning approach, specifically random forest regression, to predict optimal polling intervals.
• Implemented and evaluated NetAP-ML within a Linux environment with varying numbers of virtual machines and threads.

Main Results:

• NetAP-ML effectively shrinks the search space for optimal polling intervals.
• Achieved up to 23% higher bandwidth compared to state-of-the-art techniques.
• Demonstrated minimized performance degradation during the search process.

Conclusions:

• NetAP-ML offers a more accurate and efficient method for determining polling intervals in edge computing.
• The proposed technique significantly enhances the performance of IoT devices at the edge.