Questions on Comparative Benchmarks for RooFit in Python channels vs. bins results

👋 Hi @dawerner and @rembserj. It was nice to see Daniel give a talk at PyHEP 2023 on the benchmarking studies he did this Summer. The slides on the HistFactory benchmarks showing a large performance hit for multiple channels compared to many bins in a single channel are a bit confusing to us though as internally they should be treated on the same footing. The slides also mention a "Python loop over channels", but can you point specifically to what you're mentioning here?

As the benchmarking code is public can you additionally tell us what config was used to make Figure 5 in http://cds.cern.ch/record/2872117/? Is it the current benchmark_HistFactory/workspace_config.json or a previous revision? The repo also just mentions the master branch of ROOT was used, but can you please give us a specific commit to build against?

@alheld and us (myself, @lheinric, @gstark) have some suggestions on how to generate more realistic workspaces in the future, but we can follow up on that on GitLab in a separate issue.