Reliable Service Function Chain Deployment Method Based on Deep Reinforcement Learning.

Area of Science:

• Computer Science
• Telecommunications Engineering
• Artificial Intelligence

Background:

• Network Function Virtualization (NFV) decouples network functions from hardware.
• Service Function Chains (SFCs) sequence Virtual Network Functions (VNFs), facing reliability challenges.
• Dynamic and accurate VNF backup placement using machine learning is a significant hurdle.

Purpose of the Study:

• To propose a novel algorithm for reliable and dynamic SFC deployment.
• To address the challenge of determining optimal backup VNF locations.
• To enhance the overall quality of service for SFCs.

Main Methods:

• Developed a priority-aware deep reinforcement learning (PA-DRL) algorithm.
• Calculated SFC and node priorities based on resource capacity and network topology.
• PA-DRL determines VNF backup schemes, optimizing for delay and network load balancing.

Main Results:

• PA-DRL demonstrated significant improvements in resource utilization (36.7%), survival rate (35.1%), and load balancing (78.9%) compared to benchmark algorithms.
• Average network delay was reduced by 14.9% using the PA-DRL approach.
• The algorithm effectively improved reliability and optimization targets in dynamic SFC environments.

Conclusions:

• The PA-DRL algorithm offers an effective solution for enhancing SFC reliability and performance.
• Dynamic backup placement using deep reinforcement learning is a viable strategy for NFV environments.
• PA-DRL outperforms existing methods in key performance indicators for network services.