Dora: A Controller Provisioning Strategy in Hierarchical Domain-based Satellite Networks

TL;DR

Dora is a reinforcement learning-based controller provisioning strategy for hierarchical satellite networks that improves scalability and performance, reducing computation time significantly compared to traditional methods.

Contribution

We propose a novel three-layer hierarchical architecture and a reinforcement learning approach, Dora, for efficient controller provisioning in satellite networks.

Findings

Dora achieves 10% better provisioning quality than benchmarks.

Dora reduces computation time to 1/30-1/90 of traditional algorithms.

Experimental results validate Dora's effectiveness in SAGIN deployments.

Abstract

The rapid proliferation of satellite constellations in Space-Air-Ground Integrated Networks (SAGIN) presents significant challenges for network management. Conventional flat network architectures struggle with synchronization and data transmission across massive distributed nodes. In response, hierarchical domain-based satellite network architectures have emerged as a scalable solution, highlighting the critical importance of controller provisioning strategies. However, existing network management architectures and traditional search-based algorithms fail to generate efficient controller provisioning solutions due to limited computational resources in satellites and strict time constraints. To address these challenges, we propose a three-layer domain-based architecture that enhances both scalability and adaptability. Furthermore, we introduce Dora, a reinforcement learning-based controller provisioning strategy designed to optimize network performance while minimizing computational overhead. Our comprehensive experimental evaluation demonstrates that Dora significantly outperforms state-of-the-art benchmarks, achieving 10% improvement in controller provisioning quality while requiring only 1/30 to 1/90 of the computation time compared to traditional algorithms. These results underscore the potential of reinforcement learning approaches for efficient satellite network management in next-generation SAGIN deployments.