Space data centers schedule ai workflows efficiently despite orbital motion
Clockwork: Scheduling Compound AI Systems in Space Data Centers
Distributed, Parallel, and Cluster ComputingArtificial IntelligenceNetworking and Internet Architecture
Summary
It is hard to run complex AI tasks in space because satellites move constantly, changing which ones can process data for specific areas. The authors designed Clockwork, a system that plans where and when to run parts of AI workflows on satellites so that delays stay low and costs stay manageable. Clockwork includes two parts: one picks which satellites and modules to activate at each moment, and the other connects these snapshots over time for a smooth, cost-effective schedule. Their tests show Clockwork serves many AI requests quickly and lowers scheduling costs compared to simpler methods.
What this means in practice
- •For satellite network operators: Optimize workflow placement and plane activation in LEO constellations to meet latency goals while minimizing operational costs.
- •For cloud infrastructure teams: Adapt scheduling approaches to dynamic resource pools with moving or changing availability constraints, improving efficiency in non-terrestrial environments.
Authors
Milos Gravara, Andrija Stanisic, Stefan Nastic
Abstract
Space data centers built on LEO constellations create an opportunity to serve Compound AI workflows close to where observations are captured. Serving such a workflow for an area of interest (AoI) requires deciding which orbital planes to provision and where to place each workflow component, while orbital motion continuously changes which satellites can serve the AoI. Provisioning must therefore be planned as a sequence of decisions over time, and these decisions are coupled because changing component placements or activating new planes incurs costs. To address this, we present Clockwork, a framework for scheduling Compound AI workflows on space data centers under end-to-end latency SLOs. Clockwork has two components: i) Castor jointly selects active planes and component placements for each constellation snapshot, while ii) Pollux connects Castor's candidate provisioning states across predicted orbital evolution to construct a low-cost schedule over time. On our experimental two-shell constellation, Castor serves 81% of instances at a 300 ms SLO, while Pollux reduces mean total scheduling cost by 56% on average and by up to 63% compared to heuristic baselines.