Q-Learning-Based Path Selection for Service Function Chaining in Network Function Virtualization Environments

This paper proposes a reinforcement learning approach to optimize Service Function Chaining (SFC) in Network Function Virtualization (NFV) environments. As 5G networks introduce unprecedented complexity in service requirements and connected devices, efficient orchestration of Virtual Network Functions (VNFs) becomes critical for maintaining Quality of Service (QoS). Traditional SFC path selection methods often struggle to adapt to dynamic network conditions and resource constraints. To address these challenges, we present a Q-Learning-based algorithm that dynamically selects optimal SFC paths based on real-time resource usage (CPU and memory) and physical node location. By modeling the problem as a Markov Decision Process (MDP), our approach enables an intelligent agent to learn efficient chaining policies through interaction with the network environment. We implemented our solution in an OpenStackbased NFV testbed and compared it against random path selection across multiple performance metrics. Experimental results demonstrate significant improvements: up to 62% reduction in packet latency, 72% faster SFC construction time, and substantial enhancement in resource utilization and physical topology optimization. The findings confirm that reinforcement learning offers a promising approach to automating complex decision-making in dynamic NFV infrastructures, particularly for service chains of increasing length and complexity.