Design, Emulation, and Control of Edge Computing Systems for the Space Cloud in LEO Satellite Networks

The advent of mega-constellations in Low Earth Orbit (LEO) has transformed satellite communications, enabling global connectivity and paving the way for integrating computational services in orbit. This thesis explores the concept of a Space Cloud, in which in-orbit infrastructures provide distributed computing capabilities beyond their traditional role as communication relays. While Multi-access Edge Computing (MEC) technologies have proven effective in terrestrial networks, their adaptation to dynamic and latency-sensitive space environments raises unresolved challenges in computing resource orchestration, including the Edge Server Activation (ESA) and Edge Server Placement (ESP) problems defined in this work. We address these challenges through models, software tools, and algorithms that couple orbital propagation with real-time network emulation to evaluate control and placement strategies under realistic conditions. Central to our contributions is MeteorNet, an open-source, continuous-time platform that integrates orbital dynamics, network virtualization, and containerized execution to reproduce operational network behavior using Docker, Mininet, and Software-Defined Networking (SDN) controllers. MeteorNet dynamically adapts link states and propagation delays to satellite motion and line-of-sight conditions, generating high-fidelity synthetic datasets for training and validating learning-based algorithms in space systems, where real data are scarce. For ESA, we develop two adaptive controllers—Fuzzy Logic and Reinforcement Learning—that implement decentralized policies based on each node’s current state and historical performance data to minimize activation time while maintaining low task failure rates, outperforming Access and Bernoulli baselines across traffic regimes. For ESP, we introduce the Farthest Point Sampling (FPS) method, a topology-aware deployment strategy for space systems. We compare FPS with alternative space, terrestrial, and hybrid deployments, showing that it can reduce mean round-trip time by 15–20\% and that, at sufficient infrastructure budgets, space-based configurations can surpass terrestrial clouds. Experiments conducted with MeteorNet, together with the results obtained for the MEC resource orchestration problems, confirm that intelligent orchestration and informed placement substantially outperform reactive or naive baselines in complex, dynamic constellations. Overall, this thesis lays the groundwork for scalable, autonomous, and latency-aware edge computing in satellite constellations by unifying mixed-integer formulations, learning-based ESA controllers, topology-aware ESP strategies, and a real-time emulation framework to advance the Space Cloud paradigm.