Dynamic Controller Assignment in LEO Satellite based SDN using Multi-Agent Deep Reinforcement Learning
Hamza Mokhtar, Xiaoqiang Di, Zhengang Jiang, Mosab Hamdan
Abstract
With the rapid growth of Low Earth Orbit (LEO) satellite network applications and the accelerated expansion of the network size, software-defined networking (SDN)-based LEO satellite networks are introduced to manage resources efficiently. These networks face significant challenges due to the dynamic and time-varying nature of LEO systems, such as traffic fluctuations and imbalanced loads. Existing approaches struggle to handle efficient controller assignment in these rapidly changing topologies. To address this, we propose a novel Multi-Agent Deep Deterministic Policy Gradient (MADDPG) scheme that extends the actor-critic algorithm for distributed decision-making. The critic utilizes global information, while each actor manages local decisions, optimizing both load balancing and latency. Simulation results demonstrate that our approach significantly reduces latency and achieves better load balancing compared to state-of-the-art techniques.