Install on a HPC

The main reason to run cmf on a high performance computing cluster like Skylla is to calibrate a model hence to run the model many times. As the built-in parallelization of cmf is never as effective as a parallelization of independent runs, this installation guide never uses the built in parallelization of cmf. This guide includes the installation of spotpy a calibration and uncertanty assesment tool, developed at our institute. Spotpy is easy to use in a parallel environment using a MPI library like OpenMPI.

These instructions are valid for the HPC cluster "skylla" at the Justus-Liebig-University, Giessen. If and how this applies to your HPC environment is hard to tell. For comparison, first a description of the properties of skylla affecting the installation:

Uses GNU modules to configure environments
Has python 2.7 or 3.5 "modules" numpy available
Uses the Open Grid Engine (former Sun Grid Engine) as job queue
Runs on CentOS 6
Has OpenMPI available

Configure environment

To check modules available use:

#!sh
module av

Python 2.7

Add the follwing modules at start up:

#!sh
module initadd gcc python2 numpy GE openmpi/gcc
module add gcc python2 numpy GE openmpi/gcc

First check your python version:

#!sh
python --version

Make sure you are using python 2.7 indeed.

If the version is <=2.7.8 then get the program pip to install more packages. For higher versions, pip is already part of Python

#!sh
wget https://bootstrap.pypa.io/get-pip.py
python get-pip.py --user

In case numpy is not available as a module do

#!sh
pip install numpy --user

Python 3

Add the follwing modules at start up:

At our the super computer of the Justus Liebig University, numpy is part of the python3 module.

#!sh
module initadd gcc python3 GE openmpi/gcc
module add gcc python3 GE openmpi/gcc

Serveral "Module Commands"

If you want to switch python version or install or remove other modules, you can use the following commands:

'''Show''' all installed modules:

#!sh
module list

'''Remove modules''' you dont want anymore from modules list. E.g. for switching from python2 to python3 useing "module rm 'modulename'" (numpy is not needed anymore becaus ist included in python3)

#!sh
module rm python2
module rm numpy

you also need to '''remove the modules''' from the '''init''' (in case you put them there)[BR](BR)

So check whats in your init

#!sh
cat .bashrc

and, in case, remove

#!sh
module initrm python2
module initrm numpy

Now you got rid of python2 and able to use python3 (remember to use '''python3''' insteand of python to call python [BR](BR)

and use '''pip3''' instead of pip and dont forget –user ;))

Compile cmf

Copy the cmf source code to your user directory and unzip it in /cmf-source

#!sh
cd cmf-source
python setup.py install noopenmp --user
cd .. # Important! You cannot test your installation from the source folder

And check your cmf installation:

#!python
python
import cmf
p=cmf.project()
exit()

Optional packages

If you want to use statistics the pandas tables or create graphs, install the following packages. They are also needed by spotpy.

'''Note''': If you like the HPC to prepare graphs during the run using matplotlib, make sure to do before pylab is imported: ‘import matplotlib; matplotlib.use('Agg’)`. This prevents the creation of a GUI.

#!sh
pip install matplotlib --user
pip install scipy --user
pip install pandas --user

Adding geometry

This is only needed if you want to use the geometry support in cmf. In most cases you will not need it.

First compile and install libgeos to your local lib directory:

#!sh
wget http://download.osgeo.org/geos/geos-3.3.5.tar.bz2
tar -xjf geos-3.3.5.tar.bz2
cd geos-3.3.5
./configure --prefix=$HOME/.local
make
make install
cd ..

Now you need to point the path to that directory

#!sh
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/.local/lib
export PATH=$PATH:$HOME/.local/bin

To make this permanent (on every start up), add these lines to the file /.bash_profile. You can do that with an editor (eg. nano) or with the commands:

#!sh
echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/.local/lib" >>~/.bash_profile
echo "export PATH=$PATH:$HOME/.local/bin" >>~/.bash_profile
pip install shapely --user

Install spotpy on a HPC

To run spotpy on the HPC you need mpi4py installed. While pip install mpi4py should do, it might still not work. In case it fails, compile it by yourself (eg with adjusted version numbers):

#!sh
wget https://bitbucket.org/mpi4py/mpi4py/downloads/mpi4py-2.0.0.tar.gz
tar -xzf mpi4py-2.0.0.tar.gz
cd mpi4py-2.0.0
python setup.py install --user
cd ..

Now you can install spotpy

#!sh
pip install spotpy --user