Underworld may be installed via a number of mechanisms: Docker, pip or native install. The recommended option is using Docker as it is quick to get started.
Detailed instructions for supported HPC platforms may be found at docs/install_guides. You may also find useful usage information (on docker/hpc/compilation/other) on the Underworld blog.
We welcome feedback and support requests at our github issue tracker.
Docker is a type of lightweight virtualisation, and is the preferred method for Underworld usage on personal computers. You will first need to install Docker on your system (see Notes on Installing Docker) and then you may install Underworld via Docker. Docker can be driven from the command line, but new users may wish to use the Docker Kitematic GUI instead for ease. Simply search for ‘underworldcode/underworld2’ within Kitematic, and then click ‘CREATE’ to launch a container. You will eventually wish to modify your container settings (again through Kitematic) to enable local folder volume mapping, which will allow you to access your local drives from within your container.
For Linux users, and those who prefer the command line, the following minimal command should be sufficient to access the Underworld2 Jupyter Notebook examples:
docker run -p 8888:8888 underworldcode/underworld2
Navigate to localhost:8888 to see the notebooks. Note that you can also use particular versions of Underworld using any of the Docker image tags. For example:
docker run -p 8888:8888 underworldcode/underworld2:2.7.1b
By default (ie, if no tag is provided), docker will use the latest stable release. A list of available tags may be found on our DockerHub page. Tags can also be accessed in Kitematic via the ellipsis button on container search results.
A number of useful docker commands are provided within the Underworld cheat-sheet.
Notes on Installing Docker¶
Linux users should be able to install docker using the distribution’s standard package manager.
Windows users should note that for Windows 10 Home you should install Docker Toolbox, while for Windows 10 Professional you should install Docker Desktop.
All users on Apple OS X should use Docker Desktop (not Docker Toolbox). The Docker Toolbox edition utilised VirtualBox for virtualisation, and therefore to access any running Jupyter servers you must browse to the virtual machine address (instead of localhost). To find the VM address, you will generally execute
docker-machine ip default
but note that this will only work correctly from the Docker Quickstart Terminal.
Native installation of Underworld is now supported either directly using the Underworld build system, or via the pip Python package manager. In either case, you will need to satisfy the requirements listed below. For usage on HPC facilities you will generally need to generate a native build, although container usage via Shifter or Singularity is now available on a number of platforms (Pawsey-Magnus & TACC-stampede2 for example).
Once you have satisfied the requirements below, you will need to obtain Underworld:
git clone https://github.com/underworldcode/underworld2.git
You can then perform configuration & compilation as follows:
cd underworld/libUnderworld ./configure.py --prefix=/underworld/install/directory ./compile.py ./scons.py install
Check available configuration options using ./configure.py –help.
Once your build is complete, you will need to update your PYTHONPATH so that Python knows where to find Underworld.
(note that if you are not using the bash shell, the required command will be different.)
You can install Underworld using pip as follows:
pip3 install -v git+https://github.com/underworldcode/underworld2@development
Note that installation via pip is still experimental and may not be robust.
PETSc: PETSc can be installed via pip these days, or is usually available via platform package managers (such as apt on Ubuntu as petsc-dev). If you have PETSc installed in a non-standard location, please set the PETSC_DIR environment variable to specify the required location.
MPI & mpi4py: You will need an implementation of MPI installed on your system. Underworld is commonly used with MPICH and OpenMPI. You will also need to install the mpi4py package (via pip) which provides Python bindings to the MPI library. If non-standard, you can specify the wrapped compilers by setting the MPICC and MPICXX environment variables.
h5py: The standard h5py (installed via pip) is the recommended version for desktop usage. However, note that it will be the non-parallel enabled version, and for large parallel simulations saving/reading data may become a bottleneck, and collective IO via MPI-enabled h5py is recommended. The following command may be useful for installed MPI-enabled h5py where necessary:
CC=mpicc HDF5_MPI="ON" HDF5_DIR=/path/to/your/hdf5/install/ pip install --no-binary=h5py h5py
or alternatively you might use CC=h5pcc (if available). Note that you will also need to have a parallel HDF5 library installed. Please check the h5py site for more information. Underworld will automatically perform save()/load() operations collectively if MPI-enabled hdf5 is available.
lavavu: For rendering of visualisations, you will also need to install lavavu (via pip). Please check the lavavu page for further installation instructions.
swig: swig generates Python bindings and is a requirement. It should be installed via your system package management system (apt/yum/brew/etc), although it is straightforward to compile from source where necessary. Note that swig4 is not currently supported, and you should instead use swig3.
git: It is generally easiest to obtain the code directly using git. It should be installed via your system package management system (apt/yum/brew/etc).
libxml2-dev: This is a requirement for Underworld. The development files for libxml2 may already be available on your system, though otherwise they should be installed via your system package management system (apt/yum/brew/etc). On some system, the development packages will be named libxml2-devel.
scons: scons is the build system used by Underworld. It is a requirement. It should be installable using pip. You should use scons versions 3.0.0 or later.
numpy numpy is a requirement and should be installable using pip.
A script to run a suite of tests may be found at the top level of the project. Simply execute it to run tests:
Those using pip installation will of course need to download the repository first.