After an intensive research phase at the Distributed Systems Group at TU Wien, the LEO Trek project is coming to a close. Led by Asst. Prof. Stefan Nastic, Cynthia Marcelino, and Thomas Pusztai introduced an open-source toolkit for orchestrating, simulating, and optimizing serverless and AI workflows across the Edge, Cloud, and Space 3D Continuum.

LEO Trek created a modular ecosystem of reusable components covering simulation, scheduling, workflow optimization, state management, federated learning, and hardware-aware AI execution.

By integrating satellite nodes into the computing continuum, LEO Trek enables workflows to run where they are most effective, considering latency, energy constraints, and dynamic topology. A key motivation has been urgent computing scenarios such as disaster response, where data from drones and Earth observation satellites can be processed collaboratively in orbit and delivered faster to decision-makers.

Although the funded project is ending, the open source components remain available to researchers, platform engineers, and developers. LEO Trek establishes a foundation for future work on performance engineering and distributed AI systems across the 3D Continuum.

Core Components of LEO Trek

LEO Trek is not a monolithic system. It is a modular ecosystem of open source components that can be used independently or combined depending on the use case:

•     Stardust: a scalable simulator for the 3D Continuum. Github: https://github.com/polaris-slo-cloud/stardust-go/
•     HyperDrive: an SLO-aware serverless scheduler for the 3D Continuum.Github: https://github.com/polaris-slo-cloud/hyper-drive/
•     ChunkFunc: a workflow resource optimizer that accounts for input size and cost–performance trade-offs. Github: https://github.com/polaris-slo-cloud/chunk-func/
•     FedCCL: a Federated Learning Framework. Github: https://github.com/polaris-slo-cloud/fedccl
•     Databelt: a state management framework for serverless workflows in the 3D Continuuum. Github: https://github.com/polaris-slo-cloud/databelt/
•     Gaia: a serverless runtime for automated CPU/GPU selection for serverless AI. Github: https://github.com/polaris-slo-cloud/gaia/

Each module is self-sufficient, containing its own documentation. They can be used independently and combined as needed.

Who is it for and how does it help? LEO Trek is designed for researchers, platform engineers, and third-party developers working on distributed, serverless, and edge systems. It explains the system architecture, core components, and extension points, enabling reuse, evaluation, and further development of the open-source codebase.

How does it work? LEO Trek is composed of a set of modular components that enable simulation, scheduling, optimization, state management, and hardware acceleration in the 3D Continuum. The components can be used independently or combined, allowing both users to run simulations and evaluations and developers to extend or integrate the system. Each component is provided as a self-contained open-source repository that includes all information required to build, configure, use, and further develop the software.