HPC ODA Commons [SWR-26-003]

Abstract

HPC ODA Commons is a community-driven platform for standardizing HPC operational data analytics. HPC sites generate enormous volumes of operational data - scheduler logs, accounting records, monitoring streams - but turning that data into actionable insight is needlessly hard. Each site builds bespoke parsers, schemas, and evaluation pipelines. Results can't be compared across institutions. Promising analytics ideas stay siloed because there's no shared language for describing the data, the experiments, or the outcomes. HPC ODA Commons fixes this by establishing community-governed contracts - versioned schemas, canonical artifacts, and benchmark recipes - that make ODA workflows discoverable, reproducible, and comparable. It pairs these standards with a practical, CLI-first toolkit that lets operators and researchers go from raw logs to standardized results without sending data off-cluster.