IDAP 200 Gbps with ATLAS PHYSLITE

Targeting analysis at 200 Gbps with ATLAS PHYSLITE. This repository is very much a work in progress.

ATLAS does not have released OpenData, so there isn't an AGC we can copy and try to run. As a result, this repository's main purpose is as a facilities test:

• Run from PHYSLITE
• Load 200 Gbps off of the PHYSLITE samples
• Push all that data downstream to DASK (or similar) workers.

We have a losely tracked set of lessons learned.

Description of files

• materialize_branches.ipynb: read list of branches, distributable with Dask (use for benchmarking)

Usage

When run on the UChicago AF Jupyter Notebook no package installs are required.

There is a requirements.txt which should allow this to be run on a bare-bones machine (modulo location of files, etc.).

If you are going to use the servicex version, you have to pin dask_awkward==2024.2.0. The future versions have a bug which hasn't been fixed yet.

Input file details

The folder input_files contains the list of input containers / files and related metadata plus scripts to produce these.

In total:

• number of files: 219,029
• size: 191.073 TB
• number of events: 23,347,787,104

with additional files:

• input_files/find_containers.py: query rucio for a list of containers, given a list of (hardcoded) DSIDs

• input_files/container_list.txt: list of containers to run over

• input_files/produce_container_metadata.py: query metadata for containers: number of files / events, size

• input_files/container_metadata.json: output of input_files/produce_container_metadata.py with container metadata

• input_files/get_file_list.py: for a given dataset creates a txt file listing file access paths that include appropriate xcache. The same kind of output can be obtained by doing:

  export SITE_NAME=AF_200
  rucio list-file-replicas mc20_13TeV:mc20_13TeV.364126.Sherpa_221_NNPDF30NNLO_Zee_MAXHTPTV500_1000.deriv.DAOD_PHYSLITE.e5299_s3681_r13145_p6026 --protocol root  --pfns --rses MWT2_UC_LOCALGROUPDISK

• input_files/containers_to_files.py: process the list of containers into a list of files per container with hardcoded xcache instances, writes to input_files/file_lists/*.

Branch list determination

Branches to be read are determined with a 2018 data file.

• input_files/size_per_branch.ipynb: produce breakdown of branch sizes for given file
• input_files/branch_sizes.json: output of notebook above

Acknowledgements

This work was supported by the U.S. National Science Foundation (NSF) cooperative agreements OAC-1836650 and PHY-2323298 (IRIS-HEP).