Answers are grounded in this paper and cite their sources.
With the rapid advancement of Network Function Virtualization (NFV) and the emergence of architectures based on Mobile Edge Computing (MEC), the flexibility, scalability, and efficiency of modern networks have significantly improved. In such architectures, Virtual Network Functions (VNFs) enable dynamic and efficient service provisioning. Service Function Chains (SFCs), which consist of multiple VNFs, play a critical role in network service delivery, and their placement directly affects Quality of Service (QoS), resource utilization, and network delay. Although Deep Reinforcement Learning (DRL)-based approaches have been widely adopted to address the SFC placement problem, many existing methods fail to effectively capture the sequential dependencies among VNFs and the graph structure of network topologies. Moreover, in most existing approaches, placement decisions are generated in a VNF-by-VNF manner, which reduces efficiency and limits scalability. To address these challenges, this study proposes a Transformer-enhanced Actor–Critic Reinforcement Learning (TACRL) framework for optimizing SFC placement in MEC networks. The proposed framework is designed to maximize the SFC acceptance ratio while minimizing energy consumption, resource utilization, and service delay. By leveraging a self-attention mechanism within a transformer encoder, TACRL captures complex dependencies among VNFs and enables coordinated and parallel decision-making for the entire SFC in a single step. Unlike conventional approaches, TACRL generates the complete placement configuration in an integrated manner, thereby reducing cumulative decision errors and improving placement efficiency. In addition, an effective exploration strategy is incorporated into the DRL learning process to accelerate convergence and enhance learning performance. Simulation results demonstrate that TACRL outperforms state-of-the-art methods in achieving long-term optimization objectives and scalability. On average, the proposed framework achieves improvements of 3.2% in acceptance ratio, 3.9% in energy consumption, 5.4% in resource utilization, and 2.8% in service delay compared with baseline approaches.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: