Speaker
Description
For a few years, INFN has been developing R&D projects in the context of CMS high-rate analyses in view of the challenges of the HL-LHC phase, starting in 2030. Since then, several studies have been implemented and as a result a prototype for an analysis facility has been proposed, in compliance with CMS analysis frameworks (e.g. based on Coffea, ROOT RDataFrame), and also adopting open-source industry standards for flexibility and ease of use.
All this gave the opportunity to implement systematic studies on ROOT-based workflows at scale. In this context, we present a concrete example of an analysis exploiting ROOT RDataFrame.
Recently, new opportunities in Italy have opened up the possibility of enhancing the benchmarking approach, towards specialised resources (named HPC bubbles) hosted at INFN Padova, featuring three types of heterogeneous nodes: CPU (192 cores, 1.5TB RAM, NVMe) and GPU (4 $\times$ NVIDIA H100 80GB). Preliminary results and future plans for detailed studies on metrics such as event rate, I/O operations, energy efficiency, as well as performance comparisons between TTree and RNTuple, and finally validation with a real CMS analysis will be given.