82nd ROOT Parallelism, Performance and Programming Model Meeting

Thursday 17 Sept 2020, 16:00 → 17:00 Europe/Zurich

Online

Enric Tejedor Saavedra (CERN), Stephan Hageboeck (CERN)

Notes on the features discussed:

• Ability to have metadata (e.g. cross-section weight) that's constant across a Tree or RNT file (Slide 8)
  • Makes sense to have this native in RNT? --> Opportunity to solve this long-standing problem once and for all
    • See next point what might be needed.
  • Only add this to RDF?
    • Just file and tree name is sometimes not sufficient.
    • Request that files can be labelled / enumerated when added to RDF.
      How to do something similar for all kinds of data sources?
      Let data source define a column? How to do that efficiently?
    • Need for a callback that notifies RDF when an input {file|tree} changes?
    • DefineForDataset --> DefinePerDataset? --> DefinePerSample?
    • Note also that this kind of define should not be able to read columns with entries that change per event.
• Running RDFs in parallel
  • RDF::Run --> RDF::RunInParallel?

16:00 → 17:00 Improving the RDataFrame interface for typical cases in physics analysis

Speakers: Enrico Guiraud (EP-SFT, CERN), Stefan Wunsch (KIT - Karlsruhe Institute of Technology (DE))

RDataFrame programming model.pdf