This repository contain data and analysis script used in our ARM HPC performance and energy efficiency study:

Nikolay A. Simakov, Robert L. DeLeon, Joseph P. White, Mathew D. Jones, Thomas R. Furlani. "Are we ready for broader adoption of ARM in the HPC community: Performance and Energy Efficiency Analysis of Benchmarks and Applications Executed on High-End ARM Systems". International Conference on High Performance Computing in Asia-Pacific Region Workshops (HPCASIAWORKSHOP 2023), February 27-March 2, 2023, Singapore} https://doi.org/10.1145/3581576.3581618

The data set is in data folder:

  • ai_bench_performance.ods
  • appkernel.Rda - Rdata file contains two dataframes results_num and results_str containing metrics for test runs executed with XDMoD AKRR.
    • appkernel_results_num.csv - the above results_num dataframe in CSV format
    • appkernel_results_str.csv - the above results_str dataframe in CSV format
  • openfoam.Rda - Rdata file contains openfoam dataframe with openfoam metrics for test runs which were executed outside of XDMoD AKRR.
    • openfoam.csv - same as openfoam.Rda but in CSV format
  • resource_spec.csv - resource specifications table
  • ookami_power_openfoam.csv - power consumption for single OpenFOAM job executed on Ookami (ARM Fujitsu A64FX)
  • ubhpc_power_openfoam.csv - power consumption for single OpenFOAM job executed on UBHPC (Intel Ice Lake)

Analysis script 01_Analyze.Rmd aggregate data and write to out/<app_name>.csv. The later files combined in spreadsheet.xlsx for final table visualization with sparklines.

results_num and results_str consists of same columns the difference is that results_num contains numeric metrics like Wall Clock Time and results_str contains string metrics like App:Version. Below is the description of a columns:

  • resource - resource name
  • app - name of application kernel
  • nodes - number of cores used
  • cores - number of cores used
  • task_id - internal job id, rows with same task_id are from same batch job.
  • metric_type - type of metric: either parameter or statistics, sometime there is a fine line between them.
  • metric - metric name
  • value - metric value
  • units - metric units