SN0688 : Results from Large Scale STAR Raw Data Reconstruction on NERSC HPC

Author(s):R. Jefferson Porter, Jerome Lauret, Mustafa Mustafa, and Jan Balewski
Date:Mar. 9, 2018
File(s): STAR_Reconstruction_Project_NERSC_HPC.pdf
In 2017, STAR computing received a large over-target compute allocation from the Office of Nuclear Physics in DOE's Office of Science for data processing at the National Energy Research Scientific Computing Center, NERSC.  The allocation was granted for a demonstration project to run STAR raw data reconstruction at scale on NERSC High Performance Computing (HPC) systems.  A team of physicists and engineers from BNL and LBNL built and deployed a production pipeline, similar to pipelines previously used in STAR computing operations, tailored for processing on the NERSC Cori supercomputer.  This report provides overviews of both the project and the deployed processing framework, and describes results obtained from testing the framework at scale.  The scale tests culminated with concurrent processing on over 25,000 Cori compute cores continuously for two days at better than 95% efficiency.

Keywords:reconstruction, NERSC, Cori