Skip to content

Benchmark that performs MPI communication and computation simultaneously.

License

Notifications You must be signed in to change notification settings

mohamso/contention_benchmark

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 

Repository files navigation

An MPI Contention Benchmark

This repository contains a benchmark code that performs MPI communication and computation simultaneously. The computational kernel used in the code is the Triad kernel from the STREAM Triad benchmark originally developed by John D. McCalpin.

Build

In order to compile the system, it is sufficient to write the following on most Unix/Linux systems

icpc -O3 -openmp -std=c++11 -o main.o -c main.cc
mpicxx -o mpi_stream_ring main.o -O3 -openmp -std=c++11

Run

The input to the application is the number of bytes (message size) to communicate between the different MPI processes. For example, in order to define a message size of 4 MB, the input should be 419430 .

In order to execute the application on four nodes, using one MPI process per node on a system where SLURM is the default resource scheduler, please write:

srun --ntaks 4 ./mpi_stream_ring 4194304

License

Please see the LICENSE file.

About

Benchmark that performs MPI communication and computation simultaneously.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages