Sat-SFC: Service Function Chain Placement in Battery Supply Varying Satellite Networks

With the rapid deployment of low Earth orbit (LEO) mega-constellations, satellites are increasingly expected to provide Internet access, remote sensing, and telemetry services. Many of these tasks are resource-intensive, requiring both high computational capacity and stable communication links, which are scarce in dynamic satellite environments. Service function chaining (SFC) offers a promising paradigm by decomposing large services into sequential sub-services for distributed processing on satellites, thereby reducing ground communication overhead and improving responsiveness. However, efficient SFC placement in LEO networks remains challenging due to fast satellite mobility, intermittent connectivity, and fluctuating energy availability caused by orbital dynamics. To address these challenges, we propose Sat-SFC, a battery-aware reinforcement learning framework for SFC placement in dynamic LEO satellite networks. Sat-SFC jointly optimizes placement, scheduling, and quality-level selection while explicitly considering computational and energy constraints. The framework extends the PPO backbone with a dual-input encoder that captures both system states and resource features, and introduces a masked parallel action generation mechanism to efficiently handle the multi-discrete decision space. Furthermore, a battery-aware reward design ensures sustainable operation by avoiding resource depletion under varying energy conditions. Extensive simulations demonstrate the effectiveness of Sat-SFC. Compared with baseline algorithms-including PPO, A2C, A3C, DQN, and DD-DQN-Sat-SFC achieves up to 9% higher rewards than PPO under quality-dominant settings, doubles the utilization rate compared to Random and A3C, and maintains significantly lower latency than value-based methods. These results confirm that Sat-SFC can effectively balance the quality-latency trade-off and improve resource efficiency in LEO satellite networks.