Skip to content

A dataset focusing on the visualization and analysis of performance trends across computational kernels within the RAJA Performance Suite

License

Notifications You must be signed in to change notification settings

TauferLab/Dataset_TARPS

Repository files navigation

Trend Analysis for the RAJA Performance Suite (TARPS)

AboutPrerequisitesDependenciesSetupPublicationsCopyright and License

About

This project focuses on visualizing and analyzing performance trends across computational kernels within the RAJA Performance Suite (RAJAPerf) [1]. RAJAPerf is a benchmark suite developed at Lawrence Livermore National Laboratory (LLNL) that evaluates the performance portability of key computational kernels on high-performance computing (HPC) systems.

This repository provides tools for transforming performance data collected from RAJAPerf runs into .csv format and visualizing key performance trends. The analysis leverages data from .cali files generated by the Caliper profiler when running RAJAPerf on LLNL’s Lassen supercomputer, which features POWER9 CPUs and V100 GPUs.

The primary objectives of this project include:

  • Data Transformation to CSV: The csv_generation.ipynb notebook converts .cali files into .csv format for easier analysis and use in SBM models.
  • CSV Data Visualization: The plot_analysis.ipynb notebook generates comparative visualizations to examine performance scaling trends, identify bottlenecks through top-down metrics [2], and assess the impact of problem sizes and parallel execution strategies like the number of ranks.

Prerequisites

In order to use this package, your system should have the following installed:

  • python3
  • pip

Dependencies

The following python packages are necessary to run the notebook in this repository:

  • llnl-thicket
  • pandas
  • plotly
  • tqdm

Setup

Here is the extensive installation instructions. Please make sure the all the prerequisites are satisfied before proceeding the following steps. Make sure you are using ssh with GitHub and you have gcc compiler in your system.

  1. Clone the source code from this repository
git clone https://github.com/TauferLab/Dataset_TARPS.git
  1. Install the necessary dependencies
cd Dataset_TARPS/
pip install -r requirements.txt

Related Publications

[1] B. Bogale, I. Lumsden, D. Sukkari, D. Yokelson, S. Brinkg, O. Pearce, and M. Taufer, "Surrogate Models for Analyzing Performance Behavior of HPC Applications using the RAJA Performance Suite," in Proceedings of the International Conference on Computational Science (ICCS), Singapore, Lecture Notes in Computer Science, vol. 15906, 2025, to appear.

[2] O. Pearce, J. Burmark, R. Hornung, Befikir Bogale, I. Lumsden, M. McKinsey, D. Yokelson, D. Boehme, S. Brink, M. Taufer, and T. Scogland, “RAJA Performance Suite: Performance Portability Analysis with Caliper and Thicket,” in Proceedings of Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC24-W), Atlanta, GA, USA, 2024, doi: 10.1109/SCW63240.2024.00162.

[3] A. Yasin, “A Top-Down Method for Performance Analysis and Counters Architecture,” in Proceedings of IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), Monterey, CA, USA, 2014, doi: 10.1109/ISPASS.2014.6844459

Copyright and License

Copyright (c) 2025, Lawrence Livermore National Security, LLC.

TARPS is distributed under the terms of the Creative Commons Attribution 4.0 International Public License.

See LICENSE for more details.

Release Number: LLNL-DATA-2007855

Acknowledgments

This research was supported by Lawrence Livermore National Laboratory under LDRD project number 24-SI-005. We also acknowledge the support of the National Science Foundation (NSF) under grant number 2331152.

Full auspices and acknowledgements can be found in NOTICE.

About

A dataset focusing on the visualization and analysis of performance trends across computational kernels within the RAJA Performance Suite

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published