Deep Reinforcement Learning for Dynamic VNF Profiling and Resource Optimization

This paper presents an AI-driven framework for multi-objective Virtual Network Function (VNF) profiling, utilizing deep reinforcement learning (DRL) to optimize key resources such as CPU, memory (MEM), and link capacity (LC) in dy-namic network environments for next-generation networks. The proposed DRL model was tested in two scenarios: vFW VNF and Snort Inline Mode VNF, to evaluate its effectiveness in resource optimization. Compared to traditional reinforcement learning (RL), DRL demonstrated approximately a 10% reduction in CPU usage across both scenarios. Additionally, DRL achieved an 8% reduction in memory usage for the vFW VNF scenario and optimized LC allocation by approximately 23% for Snort Inline Mode VNF and 8% for vFW VNF. These results highlight DRL's superior adaptability and efficiency in complex, fluctuating environments, making it a promising approach for real-time VNF resource management in next-generation networks.