Stardust: A Scalable and Extensible Simulator for the 3D Continuum

Thomas Pusztai, Jan Hisberger, Cynthia Marcelino, Stefan Nastic · 2025

Read Stardust: A Scalable and Extensible Simulator for the 3D Continuum by Thomas Pusztai, Jan Hisberger, Cynthia Marcelino, Stefan Nastic (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-01513.txt.

Introduction

Thomas Pusztai, Jan Hisberger, Cynthia Marcelino, and Stefan Nastic present Stardust, a simulator for the 3D continuum of edge, cloud, and space computing. Their 2025 paper starts from the rise of low Earth orbit satellite constellations.

Workflows may soon run seamlessly across edge, cloud, and satellite nodes, but launching satellites is costly, so orchestration ideas must be tested in simulation. Existing simulators handle only small constellations unless many machines are used. Stardust can simulate mega constellations three times the size of the largest current low Earth orbit system on a single machine. It also lets users add custom components, which aids experiments with new algorithms.

Catalog record

CREATORThomas Pusztai, Jan Hisberger, Cynthia Marcelino, Stefan Nastic
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-01513

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Stardust: A Scalable and Extensible Simulator for the 3D Continuum Thomas Pusztai, Jan Hisberger, Cynthia Marcelino, Stefan Nastic Abstract Low Earth Orbit (LEO) satellite constellations are quickly being recognized as an upcoming extension of the Edge-Cloud Continuum into a 3D Continuum. Low-latency connectivity around the Earth and increasing computational power with every new satellite generation lead to a vision of workflows being seamlessly executed across Edge, Cloud, and space nodes. High launch costs for new satellites and the need to experiment with large constellations mandate the use of simulators for validating new orchestration algorithms. Unfortunately, existing simulators only allow for relatively small constellations to be simulated without scaling to a large number of host machines. In this paper, we present Stardust, a scalable and extensible simulator for the 3D Continuum. Stardust supports 1. simulating mega constellations with 3x the size of the currently largest LEO mega constellation on a single machine, 2. experimentation with custom network routing protocols through its dynamic routing mechanism, and 3. rapid testing of orchestration algorithms or software by integrating them into the simulation as SimPlugins. We evaluate Stardust in multiple simulations to show that it is more scalable than the state-of-the-art and that it can simulate a mega constellation with up to 20.6k satellites on a single machine. I Introduction Low Earth Orbit (LEO) satellites are rapidly increasing in number in recent years. As of 2024, there are more than 8,000 LEO satellites in orbit [ 1 ] , with about 7,000 of these belonging to the Starlink mega constellation [ 2 ] , which plans to grow to more than 12,000 satellites by 2028 [ 3 ] . Amazon intends to have its Kuiper mega constellation with more than 3,236 satellites [ 4 ] complete by 2029 and the FCC has called for more competition in this sector [ 2 ] . LEO mega constellations provide low latency communication between LEO and terrestrial nodes and among terrestrial nodes. For example, Starlink’s median client-LEO-Cloud round-trip latency has recently been measured to be 40-50 ms [ 5 ] . The low altitude of LEO satellites compared to geostationary satellites allows for low latency with terrestrial nodes that are directly in range. Inter-satellite laser links (ISLs) enable the creation of large orbital networks [ 6 ] , with ISL speeds demonstrated up to 100 Gbps [ 7 ] . Since LEO satellites get more computational capabilities with every new generation, several uses beyond bent pipe communication are being investigated. Processing of Earth observation (EO) data on a single satellite has been shown by ESA [ 8 ] and plans for processing data in clusters of satellites have been proposed under various names, such as Orbital Edge Computing (OEC) [ 9 , 10 ] , satellite computing [ 11 , 12 ] , or Edge-Cloud-Space 3D Continuum [ 13 , 14 ] . Possible use cases include federated learning in space [ 15 , 16 , 17 , 18 ] , EO data compression for efficient downlinking [ 19 ] , smart agriculture [ 20 ] , and disaster response [ 13 ] . The actual use of satellites varies between the ideas, from preprocessing EO data to full-fledged compute nodes that enable seamless execution of workloads across the 3D Continuum. All proposals have in common that they require ways to evaluate their designs. Since launching new satellites is expensive and the computing capabilities anticipated for the near future are not available in space yet, evaluation of LEO computing systems must be performed using simulators or emulators. Various simulators already exist, e.g., Hypatia [ 21 ] , Celestial [ 22 ] , and StarryNet [ 23 ] . While all of them compute satellite trajectories and node-to-node latencies, most simulators have a number of shortcomings. Many solutions are emulators, e.g., Celestial [ 22 ] and StarryNet [ 23 ] , which execute microVMs or containers for each node of the 3D Continuum. This has the advantage of enabling tests of real software systems, but executing one microVM or container for every node, regardless of whether it is used or not, limits the maximum infrastructure size that can be evaluated due to the resource usage. Celestial allows suspending all VMs that are not in a certain area, but it still creates a microVM for each node. Conversely, simulators, such as Hypatia [ 21 ] , do not execute nodes and are used for network simulation only, hence software testing or evaluation of placement algorithms that require node resource information are not possible. Many simulators and emulators focus on the network latencies, precompute them before the experiment, and disregard the positions of satellites during the experiment. Hence, it is often impossible to determine where a particular node is located during the experiment, although this information may be needed, e.g., for location-aware scheduling or for picking a satellite that will be in range of a certain ground station when it completes the next workload. Many solutions account only for LEO and ground station or Cloud nodes, but not for terrestrial Edge nodes, such as drones. Typically, each simulator or emulator is designed for a single purpose only, e.g., testing software under resource and network constraints or evaluating network routing algorithms. Existing solutions often lack extensibility, such as allowing custom logic to execute after every simulation step, which could, e.g., be used to add a new deployment to the experiment. In this paper, we present Stardust, a scalable and extensible open-source 1 1 1 https://github.com/polaris-slo-cloud/stardust and https://doi.org/10.5281/zenodo.15484629 simulator for the 3D Continuum. Our main contributions are: 1. Stardust, a scalable and extensible next generation simulator for the 3D Continuum with support for simulating LEO -, Cloud-, and Edge nodes in a scalable manner. Stardust enables experiments for evaluating networking and orchestration algorithms for the 3D Continuum. It supports simulating mega constellations three times the size of the currently largest constellation, with almost 7k satellites on a single machine. 2. A dynamic routing mechanism that enables experimentation with different routing mechanisms by making the ISL routing protocol and the network path computation changeable. This allows, e.g., changing the default +Grid ISL routing to a different protocol or to introduce caching or hypergraph algorithms as a replacement for Dijkstra’s algorithm to calculate node-to-node network paths. 3. SimPlugin, a plugin mechanism that serves as the integration point for custom logic that Stardust should execute at every step of the simulation. A SimPlugin has access to the complete infrastructure state and, thus, allows integrating, e.g., orchestration algorithms/software that should be evaluated using Stardust. The rest of this paper is structured as follows: Section II presents a motivating use case and requirements for a next-generation simulator for the 3D Continuum. Section III explores other 3D Continuum simulators and emulators and Section IV presents the design of the Stardust simulator. In Section V we evaluate Stardust in multiple simulations and in Section VI we conclude the paper and present future work. II Motivating Use Case & Simulator Requirements In this section, we first present a motivating disaster response scenario for the use of the 3D Continuum and, subsequently, we define requirements for an extensible next-generation simulator for the 3D Continuum. II-A Motivating Scenario While there are various use

Read the complete text