Skip to content

Latest commit

 

History

History
61 lines (39 loc) · 2.67 KB

README.md

File metadata and controls

61 lines (39 loc) · 2.67 KB

Benchmarking a Parallel System

Overview

This project focuses on benchmarking parallel systems, specifically by assessing the scalability and performance of a parallel algorithm implemented in C++. The assignment involves designing experiments, collecting data, and interpreting the results to understand the behavior of the parallel system.

Assignment Details

Task

  • Benchmark a parallel algorithm provided in the a0.cpp file.
  • Choose a parallel computer (e.g., laptop, desktop, or a node in the CCR cluster) for testing.
  • Assess the performance of the code on the chosen hardware.

Problem Description

The assignment begins with an introduction to benchmarking parallel systems and emphasizes the importance of assessing scalability and performance after implementing a parallel algorithm. The provided C++ code, contained in the archive, implements a trivial parallel algorithm.

Guidelines

  1. Compilation: Use a C++20 compiler with OpenMP support, preferably GNU g++. A provided makefile can automate the compilation process.

  2. Runtime Measurement: The code is equipped to report runtime in seconds. Edit the code to call the correct function for measurement.

  3. Parallel Function Invocation: Adjust the number of threads/cores using the following command:

    OMP_NUM_THREADS=p ./a0 your_parameters

    Replace p with the desired number of threads/cores.

  4. Experiment Design:

    • Choose between strong or weak scaling.
    • Consider the memory requirements of the code and system's main memory.
    • Ensure measurements are reliable by accounting for system jitter.
    • Plan experiments to yield useful data for interpreting results.

  5. Data Analysis:

    • Benchmark by changing the number of processors (cores) following powers of two.
    • Tabulate runtime data and plot speedup and efficiency.
    • Report hardware and software details for result reproducibility.

Project Structure

  • a0.cpp: Archive containing the C++ code for benchmarking.
  • Makefile: Automation script for compiling the code.
  • schedule.py: Python script for generating Slurm batch jobs.
  • slurm.sh: Slurm batch job for running the code.
  • run.stdout: Output from running the slurm job.
  • analyze.py: Python script for parsing the generated output.
  • A0.md: Report file summarizing the experiment, analysis, and results.

Results

The project results are summarized in the A0.md report, showcasing the scalability and performance analysis of the parallel algorithm.

License

All Rights Reserved License.