Overview

Collection of Dockerfile's suitable as the foundation for development and execution of scientific applications in high performance computing (HPC) environments. This is intended to provide stable foundations that provide a diverse collection of compiler and MPI distributions suitable for application development, data analysis and, ultimately, support research. Multiple software stack foundations lowers the bar for building and testing while generally improving application's portability, ultimately making it easier to deploy and run said application on larger supercomputer resources.

Currently the following software stacks are supported:

• GNU gfortran with OpenMPI
• Intel ifort with Intel MPI

A supplemental "development" layer is available and intended to augment the base GNU- and Intel-based images by providing additional tools suitable for interactive development, debugging and profiling. This provides things like a VNC server, terminal-based editors, GDB, run-time analysis tools (e.g. Valgrind), Linux perf, strace, and friends.

This repository is maintained with Podman in mind as privilege-less containers are required in most HPC installations. While Podman isn't typically allowed in larger supercomputing centers, Singularity is and the conversion from a Podman container image to a Singularity image is provided by a helper shell script. Docker isn't regularly tested, though all images should work just fine.

NOTE: GNU Parallel is installed in each software stack and has its citation request acknowledged. If you do not agree to terms outlined in parallel --citation (or in the FAQ), then do not use the installed version of Parallel.

Quick Start

Below are commands for building images using the GNU gfortran and OpenMPI stack. Alternate stacks can be used by substituting the directory on the base image's build (e.g. ifort-classic+intel-mpi+oneAPI_base/ for Intel ifort and MPI).

NOTE: It is recommended that images be tagged using a date-based format of YYYY.MM. Update the commands below with a year and month.

Building a base image:

$ podman build \
               --build-arg NUMBER_JOBS=16 \
               -t gfortran-openmpi:YYYY.MM \
               gfortran+openmpi/

Building a development image:

$ podman build \
               --build-arg NUMBER_JOBS=16 \
               --build-arg BASE_IMAGE=gfortran-openmpi:latest \
               -t gfortran-openmpi:YYYY.MM-dev \
               development

Running the development image with the host's ${HOME}/code/ directory mapped into the container:

$ podman run \
             --rm -it \
             -v ${HOME}/code:/code \
             gfortran-openmpi:YYYY.MM-dev

Exporting the development image to Singularity's native image format:

$ podman save -o gfortran-openmpi-YYYY.MM-dev.tar gfortran-openmpi:YYYY.MM-dev
$ singularity build -F gfortran-openmpi-YYYY.MM-dev.sif \
                      docker-archive://gfortran-openmpi-YYYY.MM-dev.tar

Published Images

Select images are published to DockerHub at gthomsen/scientific-base. The following images are currently available:

Image Name	Pull Commands	Compiler	MPI	Container OS	Dependencies	Dockerfiles
scientific-base:gfortran7.5-openmpi4.1-2022.12 scientific-base:gfortran7.5-openmpi4.1-2022.12-dev	docker pull docker.io/gthomsen/scientific-base:gfortran7.5-openmpi4.1-2022.12 docker pull docker.io/gthomsen/scientific-base:gfortran7.5-openmpi4.1-2022.12-dev	gcc/gfortran 7.5	OpenMPI 4.1.2	Ubuntu 18.04	HDF5 1.8.10, netCDF4 4.8.1 (4.5.3), FFTW 3.3.9	base, development
scientific-base:gfortran11.3-openmpi4.1-2022.12 scientific-base:gfortran11.3-openmpi4.1-2022.12-dev	docker pull docker.io/gthomsen/scientific-base:gfortran11.3-openmpi4.1-2022.12 docker pull docker.io/gthomsen/scientific-base:gfortran11.3-openmpi4.1-2022.12-dev	gcc/gfortran 11.3	OpenMPI 4.1.2	Ubuntu 22.04	HDF5 1.8.10, netCDF4 4.8.1 (4.5.3), FFTW 3.3.9	base, development
scientific-base:ifort2021.4-intelmpi2021.4-2022.12 scientific-base:ifort2021.4-intelmpi2021.4-2022.12-dev	docker pull docker.io/gthomsen/scientific-base:ifort2021.4-intelmpi2021.4-2022.12 docker pull docker.io/gthomsen/scientific-base:ifort2021.4-intelmpi2021.4-2022.12-dev	icc/ifort 2021.4	Intel MPI 2021.4	Ubuntu 18.04	HDF5 1.8.10, netCDF4 4.8.1 (4.5.3), FFTW 3.3.9	base, development
scientific-base:ifort2021.6-intelmpi2021.6-2022.12 scientific-base:ifort2021.6-intelmpi2021.6-2022.12-dev	docker pull docker.io/gthomsen/scientific-base:ifort2021.6-intelmpi2021.6-2022.12 docker pull docker.io/gthomsen/scientific-base:ifort2021.6-intelmpi2021.6-2022.12-dev	icc/ifort 2021.6	Intel MPI 2021.6	Ubuntu 20.04	HDF5 1.8.10, netCDF4 4.8.1 (4.5.3), FFTW 3.3.9	base, development

NOTE: The Intel images are large downloads (6 GB) and huge once expanded in the local registry (20-24 GB).

Documentation

Proper documentation is currently lacking, though is slowly being added as releases are made (and important details are rediscovered after being forgotten).

• Workflow: How-to for using containers for development of scientific applications
• Publishing images: Explanation of the image release process
• Notes for Intel oneAPI's images: Details for reference when revisiting how images are built on top of oneAPI images so they aren't gigantic