Online Service Deployment on Mega-LEO Satellite Constellations for End-to-End Delay Optimization

Satellite Edge Computing (SEC), which empowers satellites with computing capabilities, has been regarded as a promising paradigm for 6G alongside the development of mega-Low Earth Orbit (LEO) satellite constellations. With SEC, tasks traditionally performed on the ground can be offloaded and processed in the sky. Currently, most existing studies assume that the service entities have been deployed on satellites in advance, ignoring the selection of specific serving nodes. However, as the communication and computing resources of satellites vary in accordance with time, selecting the optimal node for deploying the service entity and migrating it at the right time will significantly affect users' quality of experience. In this regard, this article investigates the online service deployment on mega-LEO satellite constellations, taking into account the time-varying on-board resources and limited visible time. We formulate an optimization problem to maximize the number of successful deployments that meet delay requirements for each user under the long-term migration cost constraint. To solve this problem, we transform it into a Markov decision process and propose COMPOSE, an online satellite service deployment scheme based on convolution-proximal policy optimization. COMPOSE dynamically selects the optimal serving node and migrates the service entity in due course. The simulation results demonstrate the superior performance of COMPOSE in terms of delay satisfaction ratio, delay variance, and the number of migrations.

I. Introduction

Mega-low Earth Orbit (LEO) satellite constellations have gained significant attention in recent years due to the success of commercial satellite systems such as StarLink, Kuiper, and OneWeb. In the next five years, the three mentioned LEO satellite systems are expected to contain more than 44,732 satellites in total, significantly surpassing the number of satellites in middle Earth orbit constellations such as Galileo, GLONASS, and GPS [1]. With a large number of satellites, mega-LEO constellations can provide users with equal access, higher bandwidth, and reduced latency [2]. At the same time, the processing capabilities of satellites are improving, making it possible to offload computing workloads from users to the satellites [3].