Graph-based federated reinforcement learning to parallel SFC placement in mobile edge computing

Network Function Virtualization (NFV) in the context of Mobile Edge Computing (MEC) enables flexible and modular network services, where Virtual Network Functions (VNFs) are executed sequentially within Service Function Chains (SFCs). However, the sequential execution of SFCs often increases service latency for latency-sensitive applications. While Network Function Parallelism (NFP) reduces latency by enabling concurrent processing across multiple VNFs, the combined effects of SFC partitioning and resource availability have largely been overlooked. To address this gap, we propose a Graph-Based Federated Reinforcement Learning (GFRL) framework for parallel SFC placement, featuring three key capabilities: (1) capturing dependencies among VNFs using graph representations to create partially parallelized structures, (2) partitioning SFCs into multiple sub-SFCs with integrated NFP modules, and (3) deploying sub-SFCs in parallel through training a federated deep reinforcement learning model based on graph neural networks. Meanwhile, GFRL improves resource utilization and load balancing by incorporating live migration. GFRL is formulated as a Markov Decision Process (MDP) aimed at maximizing long-term cumulative rewards. Simulations across diverse scenarios demonstrate that GFRL significantly reduces end-to-end latency and improves resource utilization compared with state-of-the-art methods. Specifically, the proposed GFRL framework achieves, on average, a 4.1% reduction in end-to-end latency and a 6.2% improvement in throughput compared with existing baseline approaches.