Why it exists
ATLAS releases a portion of its data publicly, for education, outreach and research that does not require collaboration membership. Getting value from it is a different matter: open data is only useful if there is somewhere to put it and something to run against it. Downloading hundreds of gigabytes to a laptop is not an analysis environment.
The Open Data Facility closes that gap. It mirrors ATLAS Open Data in the United States, which removes a transatlantic hop for anyone working with it here, and it attaches the same kinds of working environment the Lab’s other facilities provide, so the data and the compute sit next to each other.
What is here
Storage is a dedicated petabyte of dCache, deliberately kept separate from the production grid storage pools so that public use and experiment production cannot interfere with each other. Alongside it sits one of the Lab’s large archival storage servers and a small set of worker nodes.
The working environments are the same ones the Lab runs elsewhere, pointed at open data instead of experiment data. BinderHub turns a Git repository into a running notebook session, which makes a tutorial or a paper’s analysis code something a reader can execute rather than only read. JupyterHub provides longer-lived interactive sessions. REANA runs workflows that are meant to be reproducible, capturing the software environment along with the steps.
All of this runs on RP1, the Lab’s multi-site research platform spanning the University of Chicago and Indiana University, which means the facility inherits that platform’s identity, scheduling and declarative management rather than needing its own.
Who it is for
Three audiences, with different needs. Educators and students, who need something that works in a classroom session without an account request. Outreach, where the point is that the data is genuinely the experiment’s. And researchers working on AI-era methods, for whom open data is often the only corpus that can be shared, benchmarked against and published openly.
That last group is why the facility is being built with the CLARIPHY Open Data and AI working group. Open physics data turns out to matter a great deal for machine learning research, where reproducibility and shared benchmarks depend on everyone being able to reach the same data.
Lineage
The Lab has been working on this problem for a long time. DASPOS, an NSF project with Notre Dame, Michigan and others, was an early attempt to work out what it takes to preserve not just data but the software and context needed to understand and trust it. REANA, which the Lab helped deploy and now runs, came out of the same line of thinking. The Open Data Facility is where those ideas meet actual public data and actual users.
