HPC Benchmark Automation Suite

Project Overview

This project is an automated regression testing framework designed to simulate High-Performance Computing (HPC) workflows. It uses ReFrame to orchestrate the compilation and execution of C++ scientific kernels (Matrix Multiplication) in a staged environment, ensuring reproducibility and performance verification.

Architecture

The suite follows a standard HPC "Build-Run-Analyze" lifecycle:

1. Stage: Isolates source code in a temporary sandbox.
2. Build: Compiles C++ kernels using g++ with optimization flags.
3. Run: Executes the binary and captures standard output.
4. Verify: regex-based sanity checks ensure numerical correctness.
5. Analyze: A Python aggregator parses logs and ingests GFlops metrics into a SQL backend.

Tech Stack

• Framework: ReFrame 4.x
• Languages: Python, C++, SQL, Bash
• Database: SQLite (Prototyping), extensible to InfluxDB
• System: Linux (WSL/Ubuntu)

Quick Start

1. Install Dependencies

pip install reframe-hpc

2. Run the Benchmark

reframe -c matrix_test.py -r --report-file report.json

3. Ingest Data

python3 aggregator.py report.json

4. Query Results

sqlite3 hpc_metrics.db "SELECT * FROM benchmark_runs;"

Sample Output

[SUCCESS] Ingesting: MatrixBenchmark -> GFlops = 0.37 GFlops