Satellite Constellations Enabled by RATA Optimise Energy Use across Tens of Thousands of Tasks

The increasing demand for real-time data processing from space is driving the need for efficient task allocation in distributed satellite constellations. Bharadwaj Veeravalli from the National University of Singapore, alongside his colleagues, investigates this challenge with a novel Resource-Aware Task Allocator (RATA) designed for Low Earth Orbit (LEO) to Low-Medium Earth Orbit (Low-MEO) systems. Their research employs a Single-Level Tree Network (SLTN) architecture to evaluate critical performance metrics like blocking probabilities, response times and energy consumption across extensive simulations involving tens of thousands of tasks. This work is significant because it identifies the practical limits of current systems and provides quantitative guidance for designing future satellite constellations capable of handling ever-increasing computational workloads. The team’s analysis reveals that CPU availability, not energy, is the primary factor limiting performance, and pinpoints a critical constellation size beyond which system performance rapidly degrades.

Experiments involve processing several tens of thousands of tasks to assess system behaviour under varied conditions, monitored by the Resource Allocation Task Agent (RATA) which tracks arrival rate and the availability of on-board compute, storage, bandwidth, and battery power. Experiments employed discrete-event simulations, processing tens of thousands of tasks to evaluate blocking probabilities, response times, energy consumption, and resource utilisation, meticulously monitoring parameters including task arrival rate, onboard compute, storage, bandwidth, battery availability, and the impact of eclipses on processing capabilities. Scientists developed a system that rigorously models orbital dynamics and battery charge-discharge cycles, utilising temporally sampled energy profiles during task execution, a departure from approaches assuming continuous satellite availability. This innovative approach enables the investigation of performance under varying workloads, from light to stress-inducing levels, to determine ultimate performance limits under combined influences.

The research harnessed this methodology to demonstrate pronounced non-linear scaling effects; while constellation capacity increases, blocking and delay escalate rapidly, yet energy consumption remains resilient with solar-aware scheduling. The work identifies a practical satellite count limit for baseline SLTNs, revealing that CPU availability, not energy, is the primary driver of blocking. Detailed analysis showed that a 10x increase in satellites necessitates roughly 25-30x more aggregate compute capacity due to coordination overhead and inter-satellite congestion, aligning with observations of an 81x response-time increase for a 6x constellation expansion. Furthermore, the study extends previous hierarchical clustering analyses, quantifying network scaling effects and demonstrating that average cluster size shrinks from 18 to 4.1 satellites as constellations grow, leading to coordination breakdown.

This methodology achieves a nuanced understanding of system behaviour, identifying thresholds where performance transitions from graceful degradation to collapse. The team’s cooperative allocation strategy, coupled with precise measurement of task completion within stringent latency requirements , with 99.8% of satellite-to-satellite tasks completing within 60 seconds when scheduled , provides quantitative guidance for future system design and resource management in distributed satellite networks. The research also explores partial fault tolerance through task distribution across SLTN members, limiting the impact of individual satellite failures without incurring substantial resource overhead.

LEO-MEO Constellation Performance Under Compute Load

Results demonstrate pronounced non-linear scaling effects as constellation size increases; while overall capacity expands, blocking probabilities and response times escalate rapidly. The team measured task sizes ranging from 2 to 15 GB, with compute intensities varying from 25M to 1.25B FLOP/MB, requiring between 5 and 60 seconds of processing time depending on the task’s complexity. Despite the increased complexity, energy consumption remained remarkably resilient when utilising solar-aware scheduling techniques, indicating effective power management strategies. The research establishes that constellations scaling from tens to hundreds of satellites introduce complex resource management challenges, demanding careful consideration of computational capacity. Further investigation revealed that tasks demand between 25M-125M FLOP/MB, requiring 10-60 seconds processing, directly impacting energy consumption and satellite battery recharging opportunities. Scientists recorded that a 6x expansion of the constellation resulted in an 81x increase in response time, highlighting the communication bottlenecks inherent in larger networks. Through extensive experimentation involving tens of thousands of tasks, the study characterises performance metrics including blocking probabilities, response times, energy consumption, and resource utilisation under varying traffic loads and constellation sizes. Results demonstrate that increasing constellation size does not yield linear improvements in capacity, with pronounced non-linear scaling observed across all performance indicators. Specifically, the work establishes empirically derived scaling laws revealing super-linear growth in blocking probability and response time, alongside a collapse in energy efficiency as constellation size increases.

The analysis identifies CPU availability as the primary limiting factor for task processing, rather than energy constraints, and highlights the impact of correlated, bursty traffic patterns on system performance. The authors acknowledge limitations related to the divisible load paradigm adopted in the task allocator, which may introduce computational overhead at high task arrival rates. Future research directions include exploring hierarchical SLTN clustering, inter-SLTN cooperation, multi-ground station architectures, and preemptive priority scheduling to enhance scalability and resource utilisation. Further investigation into modern multi-core processing architectures and global task pools is also suggested to alleviate CPU bottlenecks and enable task migration at larger scales. These findings offer quantitative guidance for designing and operating distributed systems in satellite constellations, establishing thresholds beyond which performance degrades significantly.