Dynamic Task Scheduling in LEO Satellites Mega-constellations: A Domain-based Perspective

• Main Authors: Fang, Zeru, Liu, Yan, Zhu, Kaige, Sun, Feng, Shi, Chenguang, Xie, Renchao, Tang, Qinqin, Huang, Tao
• Format: Conference Proceeding
• Language: English
• Subjects: computational domain; Computer architecture; Correlation; Costs; Dynamic scheduling; LEO satellites mega-constellations; Low earth orbit satellites; Processor scheduling; Satellite edge computing; Satellites; task scheduling
• ISSN: 2837-7109
• DOI: 10.1109/ICCC59590.2023.10507528

Low Earth orbit (LEO) satellite mega-constellations have become an essential part of future networks to achieve global coverage. With the significant development of edge computing, LEO satellite mega constellations are becoming capable of on-orbit computing. To address the constraints imposed by limited computing resources, satellite cooperative task scheduling has become a promising paradigm. However, most of the existing studies focus on small-scale task scheduling between neighboring nodes. They do not fully reflect the large-scale characteristics of mega constellations. Therefore, we consider a computational domain-based architecture. Based on this architecture, we investigate a scheme for solving large-scale task scheduling in LEO satellites mega-constellations, which enables intra- and inter-domain task scheduling optimization. Specifically, we use a reinforcement learning-based approach to obtain the lowest task processing cost in intra-domain scheduling. For inter-domain task scheduling, we balance the load of each computational domain from a global perspective. Then, considering the strong correlation and continuity between intra-domain scheduling and inter-domain scheduling, we propose the SAC-GALB algorithm. Numerical results verify the effectiveness and superiority of the scheme.