A TWO-LAYER ORCHESTRATION FRAMEWORK FOR QOS AND ENERGY-AWARE SATELLITE EDGE COMPUTING

The growing demand for computationally intensive services, including artificial intelligence inference and large scale Internet of Things applications, is driving the evolution of computing paradigms beyond traditional terrestrial infrastructures. In this context, Satellite Edge Computing has emerged as a promising paradigm within Space-Air-Ground Integrated Networks (SAGIN), enabling computational services to be deployed directly aboard satellites and providing global service availability even in remote areas. However, orbital computing platforms operate under stringent energy, thermal, and operational constraints, making efficient resource management a fundamental challenge. To address these issues, this thesis proposes a hierarchical resource management framework composed of two complementary decision layers aimed at jointly addressing computational provisioning and energy-aware resource orchestration. The first layer performs resource provisioning by determining the optimal computational capacity required to satisfy the aggregated workload while jointly considering Quality of Service (QoS), economic costs, and energy related aspects. To this end, each computational node is modeled as an M/G/r queue through a queueing-theoretic framework based on Markovian state modeling. Given the computational capacity identified by the first layer, the second layer performs energy-aware resource orchestration across the available computing satellites. This task is formulated as a Model Predictive Control (MPC) optimization problem that exploits predictions of energy harvesting, battery state evolution, and satellite visibility to dynamically orchestrate computational resources while balancing energy consumption among satellites, mitigating battery degradation and ensuring the long term sustainability and reliability of the orbital computing infrastructure. Simulation results demonstrate that the proposed hierarchical framework effectively balances computational demand and energy availability, achieving significant improvements in energy fairness and resource utilization compared with conventional orchestration strategies.