Adaptive SFC Management and Orchestration Based on DRL in Edge Intelligence for Computation Efficiency

Area of Science:

• Telecommunications Engineering
• Computer Networking
• Artificial Intelligence

Background:

• Network Functions Virtualization (NFV) enables flexible service deployment for Beyond 5G/6G networks by virtualizing network functions (VNFs) on edge computing. Service Function Chaining (SFC) is crucial for monitoring Internet of Things (IoT) resources, ensuring scalability and flexibility. Existing SFC reconfiguration methods struggle to meet low-latency demands for delay-sensitive applications while optimizing energy consumption.
• The integration of NFV and SFC in edge computing presents challenges in balancing resource availability, efficiency, latency, and energy saving for IoT networks.

Purpose of the Study:

• To propose a task management-aware SFC and orchestrating scheme, named GNN-PPO, for optimizing resource utilization and energy efficiency in Mobile Edge Computing (MEC) environments.
• To address the challenges of SFC reconfiguration in NFV-enabled IoT networks, particularly concerning latency and energy consumption.

Main Methods:

• Utilizing Graph Neural Networks (GNN), specifically Message Passing Neural Networks (MPNN), to model MEC node states, physical resources, and VNF capabilities for traffic fluctuation analysis.
• Employing Deep Reinforcement Learning (DRL) with Proximal Policy Gradient (PPO) for network policy determination, optimizing resource utilization and energy consumption on MEC servers.
• Developing a novel network architecture based on PPO for continuous policy enforcement and efficient resource management.

Main Results:

• The GNN-PPO scheme demonstrates significant improvements in key performance indicators compared to reference schemes.
• Experimental results confirm enhanced rewards, higher average request acceptance rates, improved SFC success rates, and better packet delivery and throughput.
• The proposed solution shows superior resource utilization ratios and confirms its scalability and practical suitability for IoT network deployments.

Conclusions:

• The GNN-PPO framework effectively optimizes resource utilization and energy consumption in NFV-enabled IoT networks through intelligent SFC orchestration.
• The integration of GNN and PPO provides a robust solution for managing complex network dynamics and meeting the stringent requirements of delay-sensitive applications.
• The proposed scheme offers a scalable and practical approach for deploying and managing future Beyond 5G/6G IoT networks.