On-Demand Centralized Resource Allocation for IoT Applications: AI-Enabled Benchmark.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Emerging technologies like IoT generate vast data, increasing demand for efficient network resource allocation.
• Generalized Processor Sharing (GPS) is vital for managing differentiated services and on-demand resource allocation in networks.
• Existing performance prediction methods for GPS often lack real-world applicability and a standardized benchmark.

Purpose of the Study:

• To address the limitations of existing Generalized Processor Sharing (GPS) performance prediction methods.
• To introduce an AI-enabled benchmark for evaluating GPS performance prediction techniques.
• To provide a comprehensive comparison of traditional and AI-based methods under various traffic conditions.

Main Methods:

• Development of an AI-enabled performance benchmark for GPS.
• Implementation and evaluation of traditional approximate analytical methods.
• Application and assessment of traditional machine learning and deep learning-based methods.
• Conducting comprehensive traffic-oriented experiments with diverse traffic flows and settings.

Main Results:

• Experimental analysis of traditional analytical, machine learning, and deep learning methods for GPS performance prediction.
• Evaluation of method and feature-level performance under different traffic conditions.
• Identification of insights for improving future GPS performance prediction.

Conclusions:

• The presented AI-enabled benchmark facilitates a fair comparison of GPS performance prediction methods.
• The study highlights the need for adaptable and comprehensive methods for real-world IoT scenarios.
• Findings offer valuable guidance for future research in network resource allocation and performance prediction.