Service-Enabled Science: OpenCosmo

As part of the ALCF's Service-Enabled Science training series, this session will explore how ALCF supports data-intensive science through large-scale data sharing and interactive HPC and AI workflows.

Efficient data sharing combined with access to computational resources is becoming essential across the sciences. Large-scale simulations on HPC platforms are now as important for scientific advances as the associated experimental and theoretical studies. Sharing these computationally expensive results enables comparison with experiments and accelerates discovery.

In this webinar, we introduce OpenCosmo, a project designed to make large cosmological simulation datasets widely available and enable flexible data access and analysis. The architecture is based on Globus services, using Globus Auth for federated identity management across facilities, Globus Flows for orchestrating multi-step workflows, and Globus Compute to execute analysis tasks on HPC resources. Users can explore data interactively through a web portal, while MCP (Model Context Protocol) servers expose analysis workflows to AI agents for natural-language-driven data discovery.

A key component is OpenCosmo's client-server model: users submit interactive queries from lightweight environments such as login nodes or ALCF Jupyter notebooks, while the server distributes execution across multiple compute nodes. This architecture decouples user interaction from compute-intensive operations, enabling responsive exploration of large datasets. The framework is designed to be extensible to other scientific domains seeking to couple data sharing with distributed computational capability.
 

Presenters
 

Patrick Wells is a cosmologist and software engineer working to develop tools for analyzing modern petabyte-scale astronomical datasets. His day-to-day work at Argonne focuses on adapting high-performance computing systems systems to provide services that enables astrophysicists  to conduct science at scale without worrying about managing data or computing resources. He is a member of the TDCOSMOcollaboration, which uses strong lensing to constrain cosmological parameters such as the Hubble constant. Patrick uses his software to study the local environments of these lenses and provide estimations of the impact of external structures on the results inferred from the strong lens itself. He also has broader interests in cosmological structure, weak lensing, and computational science.

Michael Buehlmann is a computational cosmologist with a broad interest in large-scale structure formation. His work is focused on the development and application of numerical methods to run and analyze large simulations of our Universe. Together with Nick Frontiere and JD Emberson, he is working on extending the baryonic physics models in the HACC simulation code.