Decentralized and orbit-aware scheduling for computational tasks in satellite edge computing

The rapid deployment of Low Earth Orbit (Low Earth Orbit (LEO)) mega constellations is enabling the transition from simple data relay to Satellite Edge Computing (SEC), where data processing occurs directly onboard spacecraft. However, realizing this vision requires overcoming severe challenges imposed by the harsh space environment: extreme node mobility, intermittent connectivity, and strict Size, Weight, and Power (SWaP) constraints. Traditional centralized scheduling approaches, while optimal in static networks, suffer from prohibitive signaling latency and scalability bottlenecks in LEO scenarios. Conversely, existing distributed heuristics often neglect the deterministic nature of orbital mechanics, leading to service failures when satellites lose visibility of the ground user ("sunset condition") during task execution. This thesis addresses these gaps by proposing a comprehensive framework for Decentralized, Orbit-Aware, and Energy-Efficient Service Handling. First, we introduce the Decentralized Task Scheduling (DTS) architecture, which moves decision-making locally to each Satellite Edge Node (SEN), eliminating the single point of failure and ensuring that both signaling overhead and computational complexity remain independent of the constellation size. Second, we integrate orbital dynamics into the scheduling logic. By leveraging onboard Simplified General Perturbation 4 (SGP4) propagators, we define Orbit-Aware strategies that exploit the "residual visibility time" as a hard resource constraint. We demonstrate that this predictive approach virtually eliminates mobility-induced failures for long running tasks, reducing the failure rate from ≈ 26% to less than 1%. Third, we extend the framework to handle Heterogeneous Workloads characterized by diverse computational and data-transfer requirements (e.g., CPU-intensive vs. Data intensive ) under strict energy budgets. We compare our proposed heuristics against an optimal Integer Linear Programming (ILP) baseline. Results obtained via a high fidelity discrete-event simulator show that while ILP offers theoretical energy bounds, our heuristic strategies achieve significantly higher throughput and robustness in real-time scenarios, efficiently handling the trade-off between execution speed and energy preservation. Finally, we propose a Dynamic Hybrid Routing protocol that combines topological and geometric forwarding to ensure reliable result delivery in fragmented dynamic topologies.