Configuring modelE on your machine

Before you start working with modelE you have to make sure that all necessary software is installed on your computer. This includes a Fortran compiler (gfortran or intel), netcdf library, MPI library, Git and wget.

On Mac OS computer you can install all this software with MacPorts (MacPorts have to be installed on your computer, see MacPorts installation guide for instructions). The following instructions assume that you are going to work with gfortran 4.9 compiler:

    sudo port -v install gcc49
    sudo port -v install netcdf +gcc49
    sudo port -v install netcdf-fortran +gcc49
    sudo port -v install openmpi +gcc49
    sudo port -v install git-core
    sudo port -v install wget
    sudo port select --set gcc mp-gcc49
Make sure that the directiry with MacPorts executables (/opt/local/bin) appears in your PATH before any of the system directories (this is necessary so that the model uses cpp provided by MacPorts, rather than the one which comes with Xcode and lacks GNU compatibility).

On Linux machine you can also install these with the package manager (say, with yum on CentOS), but it is advisable to compile netcdf and openmpi from the source to ensure their compatibility with the fortran compiler you are using. If you are running relatively "new" version of Linux (like Fedora 22) you can install all necessary binary packages with:

   yum install gcc.x86_64
   yum install gcc-gfortran.x86_64
   yum install gcc-c++.x86_64
   yum install git
   yum install netcdf.x86_64
   yum install netcdf-fortran.x86_64
   yum install netcdf-devel.x86_64
   yum install netcdf-fortran-devel.x86_64
   yum install openmpi.x86_64
   yum install openmpi-devel.x86_64
On Linux, if using yum package for MPI, typically you have to enable it with something like
   module load openmpi-x86_64
before you can use in in your shell window. (you may want to add this command to your ~/.profile).

On enterprise Linux platforms (such as RedHat or CentOS) the version of the default compiler may be too old to be compatible with current modelE source code. In such case you may need to install the latest Developmant Tool Set. On CentOS you can do it with

   yum install centos-release-scl
   yum install devtoolset-6
To use it in your current shell, you will need to enable it with

   scl enable devtoolset-6 bash
(make sure that your ~/.bashrc doesn't break it) or you can add

   source /opt/rh/devtoolset-6/enable
to your ~/.bashrc file, so that it is enabled automatically each time you start a new shell.

Once all necessary software is installed you can proceed to downloading the model source code:

    git clone
Now switch to "decks" directory (that's where all "make" commands should be executed) and execute "make config":

    cd modelE/decks
    make config COMPILER=gfortran ModelE_Support=$HOME/ModelE_Support
Here "$HOME/ModelE_Support" is a directory where you want modelE to store all its data (input files, administrative data, model output etc.). You can use a different path for it if you like. The "config" command will create a "ModelE_Support" directory with a necessary directory tree. It will also create a ~/.modelErc config file with your global settings. You have to edit it to set the type of compiler and MPI distribution you are using and also the location of your netcdf and MPI libraries. For the software installed with MacPorts in the example above you will need to set:


On Linux machine, typically, you would not need the last three lines (if MPI package was installed with yum). Instead, you would need to load a corresponding shell module (see above). If you have compiled MPI library from the source, typically, you would need just to set MPIDIR, pointing it to the directory where you installed your MPI package. Your computer should now be ready to run modelE jobs, both serial and parallel ones.

Other options

If you want to use a different compiler, i.e. Intel ifort you should set:


Standard installation of netcdf library assumes that the library files are located in $NETCDFHOME/lib and its include files are in $NETCDFHOME/include. If your computer uses a different setup you can specify these locations explicitly by setting the variables


ModelE supports the following MPI distributions: intel, openmpi, mpich2, mvapich2, SCALI. Set MPIDISTR variable to the name of the distribution you are using. For example if you are using Intel MPI set:


Getting it work with default yum packages on CentOS 5

You will not be able to make it work with default NetCDF package - it was compiled with gfortran compiler which is too old. The compilation will fail at the time of linking. Instead, compile NetCDF from the source as described in Appendix.

You can use MPI package which comes with CentOS 5. Both OpenMPI and MPICH2 work, though OpenMPI would produce some warning messages during the execution and is less straightforward to configure. So, if you have to use precompiled MPI package with CentOS 5, we recommend using MPICH2.

MPICH2 seems to come pre-installed on CentOS 5 machines. If not, you have to "yum install" the packages "mpich2" and "mpich2-devel". MPICH2 is controlled by Linux modules, so before using it in a shell window you have to activate it with

    module load mpich2-x86_64
(you may want to include this command into your ~/.profile or ~.bashrc). MPICH2 uses a daemon for its communications, so one has to start it before starting any MPI jobs:
    mpd &
It has to be started only once per session, so don't put it into your ~/.bashrc . Since information about MPICH2 is provided by a Linux module, you don't need to specify MPIDIR. So, variables in your ~/.modelErc should look like
Here it is assumed that you have installed your NetCDF library to /opt/netcdf/3.6.3 .