Service Orchestration Strategy in STINs: A Multi-Task SFC Mapping Approach

3GPP NTN enables deploying network functions in space. However, existing service orchestration frameworks and service function chain (SFC) mapping methods struggle to meet the flexible orchestration of diverse services in satellite-terrestrial integrated networks (STINs). To address this challenge, we propose SOFTIN, a three-plane service orchestration framework based on software-defined networking (SDN) and network function virtualization (NVF). The framework abstracts the mapping from user requirements to service functions into an orchestration workflow. Within SOFTIN, the SFC orchestration problem is formulated as an integer nonlinear programming (INLP) model, and a single SFC mapping algorithm based on double deep Q-network (DDQN) is designed to minimize deployment overhead. To enhance scalability and convergence stability while supporting service diversity and concurrency, we further propose a multi-task parallel mapping algorithm, termed multi-task graph neural network-based actor–critic with proximal policy optimization (MT-GACP). This algorithm integrates a graph neural network (GNN)-based representation learning network with a proximal policy optimization (PPO) module to mitigate convergence oscillations caused by topology changes. Moreover, multiple reinforcement learning agents manage SFC requests for multiple tasks. Finally, experiments on a Linux prototype system validate the SOFTIN framework's feasibility and orchestration capabilities. Simulation results show that the proposed algorithms achieve near-optimal performance across multiple metrics, such as placement success rate and resource utilization, and demonstrate excellent scalability in both multi-layer satellite networks (MLSN) and the Iridium network.