Adaptive actor-critic framework for efficient and reliability-aware SFC deployment in mobile edge computing

Mobile Edge Computing (MEC), empowered by Network Function Virtualization (NFV), has become a key enabling technology for supporting latency-sensitive and resource-intensive applications through Service Function Chains (SFCs). However, efficient SFC deployment remains a challenging optimization problem due to dynamic service requests, limited edge resources, heterogeneous network conditions, and the need to guarantee service reliability. Existing reinforcement learning-based approaches primarily focus on improving resource utilization and latency, while often overlooking reliability-aware deployment and adaptive decision-making in highly dynamic MEC environments. To address these challenges, this paper proposes a Reliability-Aware Adaptive Actor-Critic (RACC) framework for efficient SFC deployment in MEC environments. The proposed framework addresses the SFC deployment problem with an adaptive actor-critic reinforcement learning algorithm to jointly optimize VNF placement and resource allocation. A multi-objective objective function is formulated to jointly maximize the SFC acceptance rate and service reliability while minimizing service latency, energy consumption, resource utilization, and migration cost. To achieve this objective, RACC performs dependency-aware SFC partitioning for parallel execution, taking into account dependencies between VNFs. Furthermore, by integrating live migration and VNF reuse mechanisms, RACC improves resource allocation, maintains balanced load distribution, and enhances service reliability. Simulation results demonstrate that the proposed RACC framework consistently outperforms state-of-the-art baseline methods. Also, RACC outperformed reinforcement learning-based baseline models and reported higher expected cumulative reward values.