Dynamic Service Function Chain Deployment and Readjustment Method Based on Deep Reinforcement Learning.

Area of Science:

• Computer Science
• Networking
• Artificial Intelligence

Background:

• Software Defined Networking (SDN) and Network Functions Virtualization (NFV) enable flexible Service Function Chains (SFCs).
• Efficiently deploying SFCs dynamically to meet user demands presents significant challenges in complex network environments.

Purpose of the Study:

• To propose a novel dynamic SFC deployment and readjustment method to maximize the service request acceptance rate.
• To address the complexities of SFC deployment in NFV/SFC networks using intelligent algorithms.

Main Methods:

• Developed a dynamic SFC deployment and readjustment model based on the Markov Decision Process (MDP).
• Applied Reinforcement Learning (RL), specifically a Deep Q Network (DQN) combined with the M Shortest Path Algorithm (MQDR).
• Utilized two collaborative agents to dynamically manage SFCs, reducing action space complexity for improved training efficiency.

Main Results:

• The proposed MQDR method demonstrated a significant improvement in the request acceptance rate.
• Achieved approximately 25% higher acceptance rate compared to the standard DQN algorithm.
• Outperformed the Load Balancing Shortest Path (LBSP) algorithm by 9.3% in request acceptance.

Conclusions:

• The MQDR method effectively enhances SFC deployment and readjustment in virtualized networks.
• The RL-based approach with reduced action space offers a more efficient solution for dynamic network management.
• MQDR provides a practical solution for increasing service request acceptance in modern networks.