☰
Navigation :

Docker images

Docker images with the compiled Octopus binary are available in the container registry. Two images are available:

In addition to the latest tag, version-specific tags for Octopus releases starting with 16.0 are also provided.

Other available tags are meant for development purposes only and should generally not be used in production.

The images do not provide optimised builds for CPU (no vector instructions are used). Therefore, they are best suited for small calculations such as in a workshop or tutorial.

Using the image

For all subsequent commands we assume that the current working directory contains the inp file.

Serial execution

The default entrypoint is octopus so for a serial run it is not necessary to specify any command. The following line will run Octopus with the inp file and write output files to the current working directory on the host machine.

docker run --rm -v=$(pwd):/io --user=$(id -u):$(id -g) registry.gitlab.com/octopus-code/octopus:latest

The default working directory inside the container is /io. The option -v bind mounts the current working directory $(pwd) to /io inside the container so that the inp file is available inside the container and the output persists after the container stops.

On Linux it is necessary to map the user inside the container to the current user. Otherwise files written inside the container are owned by root and cannot be processed outside of the container. On Mac and Windows the --user option can be omitted.

Parallel execution

To use OpenMP specify OMP_NUM_THREADS, it defaults to 1 inside the container.

docker run --rm -v=$(pwd):/io --user=$(id -u):$(id -g) --env OMP_NUM_THREADS=2 registry.gitlab.com/octopus-code/octopus:latest

To use MPI a different command needs to be specified.

docker run --rm -v=$(pwd):/io --user=$(id -u):$(id -g) registry.gitlab.com/octopus-code/octopus:latest mpirun -n 4 octopus

Both options can be combined to run with MPI + OpenMP:

docker run --rm -v=$(pwd):/io --user=$(id -u):$(id -g) --env OMP_NUM_THREADS=2 registry.gitlab.com/octopus-code/octopus:latest mpirun -n 4 octopus

GPU execution

To use Octopus with GPU support use the other image and attach one or multiple GPUs to the container. The Docker runtime needs GPU support. On Linux install the NVIDIA Container Toolkit.

docker run --rm -v=$(pwd):/io --user=$(id -u):$(id -g) --gpus all registry.gitlab.com/octopus-code/octopus/gpu:latest

To efficiently use multiple GPUs run with MPI and use the same number of processes as GPUs.

Other commands

To see the version or available features use:

docker run --rm registry.gitlab.com/octopus-code/octopus:latest octopus --version

docker run --rm registry.gitlab.com/octopus-code/octopus:latest octopus --config

To get a shell inside the container use:

docker run -it --rm registry.gitlab.com/octopus-code/octopus:latest bash

Select specific Octopus versions

Octopus versions 16.0 and more recent

To use a particular Octopus version, the tag latest can be replaced by the version identifier, such as 16.0 or 16.1, for example:

docker run --rm registry.gitlab.com/octopus-code/octopus:16.0 octopus --version

Octopus versions 15.1 and older

Octopus versions 12.0 to 15.1 are available from a different container registry. To use version 12.0, for example, we can use:

docker run --rm docker.io/fangohr/octopus:12.0 octopus --version