-
Notifications
You must be signed in to change notification settings - Fork 133
installationHPC
The High Performance Computing (HPC) machines at INL have similar RAVEN install procedures, outlined below. Note that AFTER loading modules, RAVEN libraries must be installed as per a normal Linux installation (link below).
The process for any INL HPC system is as follows:
- Load modules
- Clone RAVEN
- Install libraries, giving the location of conda.
(Note, if you have not used the Linux command line before, it might be useful to read a tutorial on using it first: https://ubuntu.com/tutorials/command-line-for-beginners Learning about the ssh and the git commands will also be useful.)
Regardless of the HPC environment in which you want to install RAVEN, it should be cloned.
NOTE that sometimes modules need to be loaded before cloning can occur (for example, git must be available). Also note that git can only be used on the login node (such as lemhi1 or sawtooth1), not on the cluster compute nodes (if you are using hpc ondemand, you may need to ssh to the login node to run git commands that access github.)
git clone https://github.com/idaholab/raven.git
cd ravenThis step is optional, and you need to contact the raven developers to obtain the access first for the non-open source plugins. Contact us. The installation instruction can be found: RAVEN Plugins
module load python/3.10-mambaforge-2023-10-21
Now you can use establish_conda_env.sh script to install raven libraries, i.e.,
./scripts/establish_conda_env.sh --install --conda-defs /apps/local/mambaforge/etc/profile.d/conda.sh
When the libraries got installed, you can build raven:
./build_raven
ImportError: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.29' not found
Users can try:
export LD_LIBRARY_PATH=/path/to/miniconda3/envs/raven_libraries/lib
Note, if you have a LD_LIBRARY_PATH already defined, then you probably would instead need to use:
export LD_LIBRARY_PATH="/path/to/miniconda3/envs/raven_libraries/lib:$LD_LIBRARY_PATH"
so that it inserts the path instead of creating it.
-
Use Mamba to install RAVEN which can accelerate the installation process: ./scripts/establish_conda_env.sh --mamba --conda-defs /apps/local/mambaforge/etc/profile.d/conda.sh
-
Use “pip” to install RAVEN (https://pypi.org/project/raven-framework/) pip install raven-framework These should work on Linux, Mac OS and Windows. After installation, the raven_framework command can be used to run raven.
Lemhi now has a raven module that can be loaded into your HPC environment. NOTE: This may not be a cutting-edge version of RAVEN.
module load raven
This module also comes with RAVEN plugins: HERON and TEAL.
Now to run raven on an input file, you will type the command:
raven_framework <raven_input.xml>
NOTE: Ensure you do not already have a local conda environment called raven_libraries this will conflict with the module environment of the same name.
After the appropriate modules are loaded, continue with normal Linux installation instructions:
Linux installation instructions.
Note, that if you are installing RAVEN as an administrator, do not use the default raven libraries name (this will complicate other people using the default name).
Either set the name with an environment variable before running establish_conda_env.sh:
export RAVEN_LIBS_NAME=raven_libraries_hpc_raven_2_1
or edit the .ravenrc file (note that will require establish_conda_evn.sh to be rerun:
RAVEN_LIBS_NAME = raven_libraries_hpc_raven_2_1
(This technique can also be used if multiple versions of RAVEN need to be installed.)
In general RAVEN needs:
- a GCC compiler with C++11 standard (for example 4.9.2)
- conda (local installation per user is fine)
- MPI (if distributed simulations are desired)
Once these three are available, install RAVEN via the Linux installation instructions.
See also Advanced Installation