This repository contains some scripts, methods and results to compare benchmarks between containers type.
To put your own results, fork this repository and ask for pull request.
- Please create a subdirectory in method_and_results with your CPU or GPU type and create a detailed markdown file by type of container and distro to be able to rerun your benchs.
- If you made heterogeneous tests with MPI, create a subdirectory in method_and_results with your mpi version and with every details (operations/commands, latency, bandwith, CPU types on nodes...) of your runs.
- Rename the new markdown file(s) with <container_type>_<distro>_<type_of_test>.md with <type_of_test> = [single_core or mpi or openmp or cuda] e.g. singularity_centos6_mpi.md
- In the markdown file, you have to inform every details to be able to rerun your benchs
- For every runs, please create a markdown file with bare-metal results and name it host_<distro>_<type_of_test>.
- For single core tests, please use this file. It is linpacknew.c with a timeout at 2mn in a non-interactive mode.
- If you want another array size, please create a subdirectory in linpack and put the modified file you used.
- For docker and singularity, please create a Dockerfile or a singularity definition file for bootstrap (if you do not use those supplied). For lxd, inform us with the commands you used by filling the markdown file you created before.