OrbitChain: Orchestrating In-orbit Real-time Analytics of Earth Observation Data

TL;DR

Real-time Earth observation analytics are hampered by long satellite-ground contact windows and limited downlink bandwidth. OrbitChain tackles this by orchestrating orbital-edge computing resources across a constellation, modeling analytics as a sensing-and-analytics pipeline and solving function deployment via a mixed-integer program that allocates CPU quotas $r_{ ext{cpu},i,j}$ and GPU time slices $t_{i,j}$ within per-frame deadline $\\Delta_f$, while handling workloads $N_{f,i}$. It introduces an analytics application graph (a DAG) and a realization-graph-based traffic routing with a greedy algorithm to minimize inter-satellite hops. Hardware-in-the-loop experiments on orbital-edge devices show up to 60% higher analytics throughput and up to 72% lower inter-satellite traffic, delivering results within minutes under realistic bandwidths. The work enables real-time time-sensitive applications and paves the way for inter-constellation tip-and-cue collaborations.

Abstract

Earth observation analytics have the potential to serve many time-sensitive applications. However, due to limited bandwidth and duration of ground-satellite connections, it takes hours or even days to download and analyze data from existing Earth observation satellites, making real-time demands like timely disaster response impossible. Toward real-time analytics, we introduce OrbitChain, a collaborative analytics framework that orchestrates computational resources across multiple satellites in an Earth observation constellation. OrbitChain decomposes analytics applications into microservices and allocates computational resources for time-constrained analysis. A traffic routing algorithm is devised to minimize the inter-satellite communication overhead. OrbitChain adopts a pipeline workflow that completes Earth observation tasks in real-time, facilitates time-sensitive applications and inter-constellation collaborations such as tip-and-cue. To evaluate OrbitChain, we implement a hardware-in-the-loop orbital computing testbed. Experiments show that our system can complete up to 60% analytics workload than existing Earth observation analytics framework while reducing the communication overhead by up to 72%.

Figures (17)

• Figure 1: The modules and data flow in an Earth observation application for farmland flood monitoring.
• Figure 2: Constellation organization.
• Figure 3: \ref{['fig:data_parallel:visualization']} Data parallelism. \ref{['fig:data_parallel:utilization']} Memory utilization of analytics applications with varying complexity. The grey line is the device's memory capacity. Virtual memory is enabled to allow memory usage to exceed onboard limits at the cost of degraded performance. (D: cloud detection; L: landuse classification; R: crop monitoring; W: waterbody monitoring).
• Figure 4: \ref{['fig:computational_parallel:visualization']} Compute parallelism. \ref{['fig:computational_parallel:resource_contention']} Cloud detection module's inference latency, when cohosted with other modules on the same satellite. Bar heights and error bars indicate the average and standard deviation among 10-round evaluations. We refer labels to Fig. \ref{['fig:data_parallel']}.
• Figure 5: Application and realization graphs.