Adaptive Service Function Chain Migration in Satellite–Terrestrial Integrated Networks

In satellite-terrestrial integrated networks (STINs), dynamically migrating service function chains (SFCs) across heterogeneous nodes is essential to maintain quality of service under changing link and node conditions. Conventional heuristics or static optimization either incur excessive migration overhead or fail to adapt to network volatility. In this paper, we propose DWAR (Dynamic Weighting–based Adaptive Reinforcement), a Proximal Policy Optimization (PPO)–driven framework that incorporates a self-attention module to dynamically adjust multi-objective reward weights—covering end-to-end delay, migration cost, energy consumption, and load balancing—based on real-time network state. We evaluate DWAR against four baselines (ILP-based SOLVER, MSDF, PAVM, and GREEDY) in a 1000-node STIN simulator under four stress scenarios: baseline, link degradation, node overload, and traffic burst. Results show that DWAR achieves 90–95% of the optimal SOLVER delay while reducing migration cost and energy by 15–20%, and maintaining robust load balance across all scenarios. These findings demonstrate DWAR’s practical efficacy for real-time SFC management in dynamic, large-scale STINs.