Reliability-Aware Satellite Edge Computing: From Latency Modeling to System Design

Abstract

This paper analyzes the round-trip (RT) latency distribution in a cooperative and distributed satellite edge-computing system and uses it to support reliability-aware system design. In the considered architecture, a serving satellite either processes a task locally or cooperates with a processing-capable satellite through multi-hop inter-satellite links (ISLs). The resulting RT latency is jointly determined by access-link reliability, inter-satellite forwarding, and distributed onboard processing. The proposed model accounts for finite-blocklength (FBL) uplink (UL) retransmissions, stochastic local and remote processing times, asymmetric forward and return ISL delays, and offloading decisions that depend on where the input data are available. By combining these components, the developed framework can be used to compute the probability that the RT latency meets a target deadline. This enables determining the required onboard processing capability, the maximum feasible offloading range in terms of the total ISL hop count, and the energy-aware balance between local and offloaded execution under probabilistic latency constraints. The results show that, under a 99% latency-reliability requirement, offloading to a more capable remote processor reduces the required operating frequency by up to 25% compared with local execution for ISL routes between two and ten total hops, whereas offloading with identical local and remote processors increases the required frequency by up to 33% due to the additional ISL forwarding delay.