Authors: Kadidia Konate (Lawrence Berkeley National Laboratory (LBNL)), Sergio Iserte (Barcelona Supercomputing Center (BSC)), Kevin Menear (National Laboratory of the Rockies (NLR)), Estela Suarez (Jülich Supercomputing Centre, University of Bonn), Ahmad Maroof Karimi (Oak Ridge National Laboratory (ORNL))
Abstract: HPC facilities each build operational data analytics in isolation, with no shared benchmarks to compare methods or reproduce results across sites. This seventh-edition BoF convenes the community around a common goal: shared standards, open benchmarks, and cross-facility collaboration for HPC workload analysis. Using the HPC ODA Commons, introduced to the community at ISC 2026, as a working exemplar, we examine how standardized schemas and benchmark recipes enable reproducible comparison across systems. We present an emerging multi-facility study, invite sites to contribute data, schemas, or models, and work together to prioritize analyses and plan a community report.
Long Description: HPC facilities generate vast operational data, including scheduler logs, accounting records, and telemetry, yet each site develops bespoke schemas, parsers, and evaluation pipelines. Results rarely transfer: a model or metric validated at one facility cannot be easily reproduced or compared at another. Progress in operational data analytics (ODA) stays siloed, and the community lacks the shared benchmarks that transformed fields like computer vision and natural language processing.
This BoF brings facilities together around three commitments: common data standards, open benchmarks, and privacy-preserving cross-site collaboration. Rather than promoting any single tool, the session treats standardization and reproducibility as community infrastructure that no facility can build alone. We focus on the practices, such as schema conventions, benchmark recipes, and evaluation protocols, that let independent sites compare methods honestly.
To ground the discussion, we draw on the HPC ODA Commons, an open-source effort introduced to the community at ISC 2026. It offers a concrete picture of what shared standards can look like in practice: versioned schemas, recipe-driven benchmarks, and local-first tooling that keeps sensitive logs on-cluster while still allowing sanitized, derived artifacts to be shared. We present it as one model among several and invite the community to help shape the tool while it is still in its early stages.
Momentum from ISC 2026 catalyzed an early, multi-institution effort to apply consistent methods across systems from DOE laboratories, European centers, and academic partners. The study asks foundational questions: How do workload characteristics differ across facilities and architectures? Can anomaly-detection or power-prediction models trained at one site generalize to another? Which operational metrics best predict system health? This BoF is where the study's design, earliest observations, and open questions are shaped, collaboratively and in public.
Goals for SC26
(1) Align the community on standards and benchmarks for HPC workload analysis. (2) Share lessons on data standardization and privacy-preserving collaboration. (3) Expand participation across facilities and regions. (4) Gather input on priority analyses and metrics. (5) Chart a path toward a community report for funders and program managers.
Relevance
This session tackles the barriers that have limited ODA progress: absent benchmarks, non-reproducible results, and isolated development. By demonstrating that facilities can collaborate across institutional and national boundaries, it establishes a repeatable model for community-driven research and connects to broader efforts, including the EEHPCWG ODA Working Group.
Expected Outcomes
Attendees will understand how shared standards enable cross-site reproducibility, learn practical and privacy-preserving ways to contribute data or models, connect with a growing community of practice, and help set priorities for the study and the planned community report.
Track Record
This series has run at six prior venues (CUG 2024, ISC 2024, SC 2024, ISC 2025, SC 2025, ISC 2026), consistently drawing 100+ participants from national laboratories, European and Asian centers, academia, and industry. It has evolved from community-building toward shared infrastructure and coordinated study.