Overview

You can find detailed documentation at software.llnl.gov/benchpark. Benchpark can also be found on GitHub.

Benchpark is an open collaborative repository for reproducible specifications of HPC benchmarks. Benchpark enables cross-site collaboration on benchmarking by providing a mechanism for sharing reproducible, working specifications for the following:

1. System Specifications (benchmark and experiment agnostic)

• Hardware information

• System software environment information (available compilers, MPI)

• Scheduler and launcher

2. Benchmark Specifications (system and experiment agnostic)

• Source repo and version

• Build configuration (with Spack)

• Run configuration (with Ramble)

3. Experiment Specifications (specific benchmark experiment on a system specification)

• Programming model (e.g., OpenMP, CUDA, ROCm) for the benchmark on a given system

• Parameters for individual runs in a study

Dependencies

Benchpark uses the following open source projects for specifying configurations:

• Spack for building benchmark and dependencies

• Ramble for run configurations

Community

Benchpark is an open source project. Questions, discussion,and contributions of new specifications as well as updates and improvements to existing specifications are welcome.

• Adding a System Specification

• Adding a Benchmark Specification

• Adding an Experiment Specification

We use github discussions for Q&A and discussion.

Contributing

To contribute to Benchpark, please open a pull request to the develop branch. Your PR must pass Benchpark’s unit tests, and must be PEP 8 compliant.

Authors and citations

Many thanks to Benchpark’s contributors.

Benchpark was created by Olga Pearce, Alec Scott, Greg Becker, Riyaz Haque, and Nathan Hanford.

To cite Benchpark, please use the following citation:

Olga Pearce, Alec Scott, Gregory Becker, Riyaz Haque, Nathan Hanford, Stephanie Brink, Doug Jacobsen, Heidi Poxon, Jens Domke, and Todd Gamblin. 2023. Towards Collaborative Continuous Benchmarking for HPC. In Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis (SC-W 2023), November 12–17, 2023, Denver, CO, USA. ACM, New York, NY, USA, 9 pages. doi.org/10.1145/3624062.3624135.

License

Benchpark is released under the Apache 2.0 w/ LLVM Exception license. For more details see the LICENSE file.

LLNL-CODE-850629

Basics

• Getting Started
• Basic Usage
• Benchpark Workflow
• Frequently Asked Questions

Catalogue

• System Specifications
• Benchmarks and Experiments

Tutorials

• Hello Benchpark Example
• Running on an LLNL System

Running Benchpark

• Setting Up a Benchpark Workspace
• Defining the Experiments in Benchpark
• Running an Experiment in Benchpark
• Analyzing Experiments in Benchpark
• Benchpark Modifiers
• Benchpark Suites of Experiments

Contributing

• Adding a System
• Adding a Benchmark
• Adding an Experiment
• Adding a Dryrun

Legacy Workflow

• Legacy Benchpark Workflow