DVNE-DRL: dynamic virtual network embedding algorithm based on deep reinforcement learning.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Virtual network embedding (VNE) faces challenges balancing immediate decisions with long-term revenue goals.
• Existing methods often overlook the dynamic nature of virtual networks and fail to adapt to real-time network states.

Purpose of the Study:

• To develop a dynamic network embedding framework that accounts for changing user numbers and workloads.
• To propose a deep reinforcement learning-based algorithm for efficient and adaptive VNE.

Main Methods:

• Modeled VNE as a Markov Decision Process (MDP).
• Introduced a dynamic virtual network embedding algorithm (DVNE-DRL) leveraging deep learning for state perception and reinforcement learning for decision-making.
• Enhanced feature extraction and matrix optimization techniques.

Main Results:

• DVNE-DRL demonstrated improved performance compared to existing algorithms.
• Achieved approximately 25% increase in acceptance rate.
• Achieved approximately 35% increase in average revenue.

Conclusions:

• The proposed DVNE-DRL algorithm effectively addresses the dynamic nature of VNE.
• Deep reinforcement learning offers a robust solution for optimizing network resource management and revenue in complex network environments.