Trabant: A Serverless Architecture for Multi-Tenant Orbital Edge Computing

TL;DR

The paper addresses the rigidity and high upfront costs of current orbital edge computing by proposing Trabant, a serverless, time-shifted FaaS platform that enables multi-tenant on-board processing on shared Earth observation satellites. It presents the Trabant architecture (frame buffer, preprocessing, executor, and isolated tenant functions) and formalizes resource constraints with equations like $P_{\text{generated}} \geq P_{\text{compute}} + P_{\text{comm}}$, $B_{\text{downlink}}$, and latency bound $R_{\text{frame}}^{-1} \geq T_{\text{pre}} + \sum T_{f_i}(1 - R_{\text{filter}})$, then validates the approach via a Go-based prototype on a Raspberry Pi using real BUPT-1 traces and TF-Lite workloads. The evaluation demonstrates that four ML functions can be scheduled under energy/thermal budgets, shows end-to-end time-shifted processing and SEU resilience, and analyzes deployment-size trade-offs for multi-tenant deployment. The results suggest substantial reductions in mission-planning overhead and greater flexibility for Earth observation campaigns, with open-source artifacts to enable broader adoption. The work lays a foundation for scaling shared OEC platforms, discusses trust and isolation concerns, and outlines directions for constellation-scale deployment and advanced sandboxing in future work.

Abstract

Orbital edge computing reduces the data transmission needs of Earth observation satellites by processing sensor data on-board, allowing near-real-time insights while minimizing downlink costs. However, current orbital edge computing architectures are inflexible, requiring custom mission planning and high upfront development costs. In this paper, we propose a novel approach: shared Earth observation satellites that are operated by a central provider but used by multiple tenants. Each tenant can execute their own logic on-board the satellite to filter, prioritize, and analyze sensor data. We introduce Trabant, a serverless architecture for shared satellite platforms, leveraging the Function-as-a-Service (FaaS) paradigm and time-shifted computing. This architecture abstracts operational complexities, enabling dynamic scheduling under satellite resource constraints, reducing deployment overhead, and aligning event-driven satellite observations with intermittent computation. We present the design of Trabant, demonstrate its capabilities with a proof-of-concept prototype, and evaluate it using real satellite computing telemetry data. Our findings suggest that Trabant can significantly reduce mission planning overheads, offering a scalable and efficient platform for diverse Earth observation missions.

Figures (11)

• Figure 1: A shared satellite with orbital edge computing could continuously capture Earth observation data and process it using services and models from different tenants, reducing the downlink strain while still providing near-real-time insights
• Figure 2: Traditional Earth observation satellite capture images of Earth from LEO and downlink them when passing ground stations, which are usually located near the Earth's poles. Data is sent through the backhaul network to a cloud for processing, from which scientists can access it for analysis
• Figure 3: Trabant is a serverless architecture for on-board processing of Earth observation frames. Captured frames are preprocessed and prefiltered, with an executor either directly invoking tenants' functions or enqueuing them when temperature or power constraints do not allow processing. The function output, which can be filtered or inferred data, is then buffered for downlinking.
• Figure 4: During our 6-hour trace, BUPT-1 consumes a mean 13.51 of power, while its solar array generates between 0.00 (darkness) and 48.59 (maximum at sunlight) of power. Power output is also unstable during sunlit periods, with more power generated at the beginning than at the end, a result of the satellite's angle relative to the sun.
• Figure 5: Our 6-hour trace from May 1st, 2023, encompasses four orbits, with the satellite coming into contact with the ground station in Tongchuan toward the end of its third orbit