Getting simform running the first time can be très-annoying! On ubuntu, with
CDH3 on EC2, it's actually pretty painless -- see the demo directory.
But no one who puts up real clusters seems to use Ubuntu.
All nodes
- A recent branch of python2, say 2.6
- numpy
- pynetcdf or python-netcdf (Scientific.IO.NetCDF)
- Hadoop -- tested with CDH3, earlier prototypes tested with Hadoop 0.21
- python-hadoop package
Only the head node, or job submission node
- make -- nothing special here, any version should do
- dumbo
- dumbo-feathers -- this requires compiling some Java against the current version of Hadoop
- A version of the JDK to compile dumbo-feathers
- mrjob
Recommended for all nodes
- ctypedbytes -- this makes all the IO go faster
Recommended for the head node
- git
One of the most common problems is forgetting to set the HADOOP_HOME
environment variable. This should be set to whereever your hadoop
installation lives, e.g. /usr/lib/hadoop
See the demo directory.
All nodes
sudo apt-get install python-dev python-numpy \
python-setuptools python-netcdf \
libatlast3gf-base
sudo easy_install -z typedbytes
sudo easy_install ctypedbytes
Head node
Install the easy stuff
sudo apt-get install git-core
sudo easy_install -z dumbo
Install MRJob
mkdir ~/devextern
cd ~/devextern
git clone https://github.com/dgleich/mrjob.git
cd mrjob
sudo python setup.py install
Install python-hadoop locally
cd ~/devextern
git clone https://github.com/hyysun/Hadoop.git
cd Hadoop/python-hadoop
python setup.py sdist
sudo easy_install dist/Hadoop-0.20.tar.gz
You must install the same Hadoop-0.20.tar.gz
file via
easy_install
on all nodes -- see below. We may be able to turn
this into an egg distribution in the future.
Install feathers
You must install feathers into a subdirectory of the current simform distribution. Here's an example:
cd simform/src # THIS ONLY WORKS AFTER YOU INSTALL SIMFORM!
cd model
git clone https://github.com/klbostee/feathers.git
cd feathers
sh build.sh
cp feathers.jar ..
All nodes again
<copy Hadoop-0.20.tar.gz from head>
sudo easy_install Hadoop-0.20.tar.gz
For these last steps, here is what I did on a test cluster for EC2:
for node in 10.34.86.56 10.76.213.246 10.116.190.35 10.35.94.101 10.113.42.21; do scp -o "UserKnownHostsFile /dev/null" -o StrictHostKeyChecking=no dist/Hadoop-0.2.tar.gz $node:~ ssh -o "UserKnownHostsFile /dev/null" -o StrictHostKeyChecking=no $node sudo easy_install Hadoop-0.2.tar.gz done
These were written for a different purpose, but do contain all of the install instructions. If you think it's easier to follow the ubuntu instructions, you can follow them for any of the install except for the apt-get pieces, which, obviously, don't work on Redhat.
Here are some install instructions you can reference to meet the requirements:
On all nodes (hadoop namenode and datanodes), we should install the following packages:
# install numpy
wget "http://sourceforge.net/projects/numpy/files/NumPy/1.6.1/numpy-1.6.1.tar.gz/download"
tar xfz numpy-1.6.1.tar.gz
cd numpy-1.6.1
python setup.py install
cd ..
# install ctypedbytes
git clone https://github.com/klbostee/ctypedbytes.git
cd ctypedbytes
python setup.py install
cd ..
# install netcdf
wget "http://www.unidata.ucar.edu/downloads/netcdf/ftp/netcdf-4.1.3.tar.gz"
tar xfz netcdf-4.1.3.tar.gz
cd netcdf-4.1.3
./configure --with-pic --disable-dap --disable-netcdf-4
make
make install
make check
cd ..
# install pynetcdf
wget "http://downloads.sourceforge.net/project/pylab/PyNetCDF/0.7/pynetcdf-0.7.tar.gz"
tar xfz pynetcdf-0.7.tar.gz
cd pynetcdf-0.7
python setup.py install
cd ..
# install hyy-python-hadoop
git clone https://github.com/hyysun/Hadoop.git
cd Hadoop
cd pyhton-hadoop
python setup.py install
cd ..
cd ..
If you have some problems about netcdf, try:
- Place usr-local.conf in /etc/ld.so.conf.d on all nodes. (see icme-hadoop1 /etc/ld.so.conf.d/ for the template)
- Run ldconfig -v as root on all nodes.
We need to install the following packages on the Hadoop namenode:
# install mrjob
git clone https://github.com/dgleich/mrjob.git
cd mrjob
python setup.py install
cd ..
# install dumbo
git clone https://github.com/klbostee/dumbo
cd dumbo
python setup.py install
cd ..
- To get dumbo feathers, run following: git clone https://github.com/klbostee/feathers.git cd feathers bash build.sh
After that, move feathers.jar into your java classpath or just copy feathers.jar to src/model.