Resource-aware dynamic scheduling for energy-efficient LEO constellation-as-a-service.

Abstract

Low Earth Orbit (LEO) Constellation-as-a-Service (LEO-CaaS) is emerging as a distributed space-edge computing paradigm. However, satellite mobility makes connectivity and onboard resources time-varying, complicating energy-efficient orchestration for latency-constrained services. We capture these coupled dynamics by formulating joint task scheduling and resource allocation as a mixed-integer nonlinear program that minimizes system energy under latency, visibility, contact window, communication, and computation constraints. To enable online scheduling, we propose Computation and Resource Aware Fair Task Scheduling (CRAFTS), a low-complexity two-stage heuristic that combines workload-aware admission, resource-contention-aware satellite ranking, and workload-proportional redistribution. Simulation results under dynamic LEO-CaaS conditions show that CRAFTS reduces total energy by up to 29% compared with a multi-agent learning benchmark while achieving sub-second scheduling. It also consistently lowers energy per assigned task over heuristic baselines, demonstrating scalable online scheduling performance.