2. Install and Build the HPC-Stack

Attention

The HPC-Stack is already installed on Level 1 systems (e.g., Cheyenne, Hera, Orion). Installation is not necessary.

HPC-Stack installation will vary from system to system because there are so many possible combinations of operating systems, compilers, MPI’s, and package versions. Installation via an EPIC-provided container is recommended to reduce this variability. However, users may choose a non-container approach to installation if they prefer.

Note

MPI stands for Message Passing Interface. An MPI is a standardized communication system used in parallel programming. It establishes portable and efficient syntax for the exchange of messages and data between multiple processors that are used by a single computer program. An MPI is required for high-performance computing (HPC).

2.1. Install and Build the HPC-Stack in a Singularity Container

The Earth Prediction Innovation Center (EPIC) provides several containers available for the installation of the HPC-Stack either individually or combined with Unified Forecast System (UFS) applications:

2.1.1. Install Singularity

To install the HPC-Stack via Singularity container, first install the Singularity package according to the Singularity Installation Guide. This will include the installation of dependencies and the installation of the Go programming language. SingularityCE Version 3.7 or above is recommended.

Warning

Docker containers can only be run with root privileges, and users cannot have root privileges on HPC’s. Therefore, it is not possible to build the HPC-Stack inside a Docker container on an HPC system. A Docker image may be pulled, but it must be run inside a container such as Singularity. Docker can, however, be used to build the HPC-Stack on a local system.

2.1.2. Build and Run the Container

  1. Pull and build the container.

    singularity pull ubuntu20.04-gnu9.3.sif docker://noaaepic/ubuntu20.04-gnu9.3
    singularity build --sandbox ubuntu20.04-gnu9.3 ubuntu20.04-gnu9.3.sif
    cd ubuntu20.04-gnu9.3
    

    Make a directory (e.g., contrib) in the container if one does not exist:

    mkdir contrib
    cd ..
    
  2. From the local working directory, start the container and run an interactive shell within it. This command also binds the local working directory to the container so that data can be shared between them.

    singularity shell -e --writable --bind /<local_dir>:/contrib ubuntu20.04-gnu9.3
    

    Make sure to update <local_dir> with the name of your local working directory.

2.1.3. Build the HPC-Stack

  1. Clone the HPC-Stack repository (from inside the Singularity shell initialized above).

    git clone https://github.com/NOAA-EMC/hpc-stack
    cd hpc-stack
    
  2. Set up the build environment. Be sure to change the prefix argument in the code below to your system’s install location (likely within the hpc-stack directory).

    ./setup_modules.sh -p <prefix> -c config/config_custom.sh
    

    Here, <prefix> is the directory where the software packages will be installed with a default value of $HOME/opt. For example, if the HPC-Stack is installed in the user’s directory, the prefix might be /home/$USER/hpc-stack/hpc-modules.

    Enter YES/YES/YES when the option is presented. Then modify build_stack.sh with the following commands:

    sed -i "10 a source /usr/share/lmod/6.6/init/bash" ./build_stack.sh
    sed -i "10 a export PATH=/usr/local/sbin:/usr/local/bin:$PATH" ./build_stack.sh
    sed -i "10 a export LD_LIBRARY_PATH=/usr/local/lib64:/usr/local/lib:$LD_LIBRARY_PATH" ./build_stack.sh
    
  3. Build the environment. This may take up to a couple of hours to complete.

    ./build_stack.sh -p <prefix> -c config/config_custom.sh -y stack/stack_custom.yaml -m
    
  4. Load the required modules, making sure to change the <prefix> to the location of the module files.

    source /usr/share/lmod/lmod/init/bash
    module use <prefix>/hpc-modules/modulefiles/stack
    module load hpc hpc-gnu hpc-openmpi
    module avail
    

From here, the user can continue to install and run applications that depend on the HPC-Stack, such as the UFS Short Range Weather (SRW) Application.

2.2. Non-Container HPC-Stack Installation and Build (General/Linux)

2.2.1. Install Prerequisites

To install the HPC-Stack locally, the following pre-requisites must be installed:

  • Python 3: Can be obtained either from the main distributor or from Anaconda.

  • Compilers: Distributions of Fortran, C, and C++ compilers that work for your system.

  • Message Passing Interface (MPI) libraries for multi-processor and multi-core communications, configured to work with your corresponding Fortran, C, and C++ compilers.

  • Programs and software packages: Lmod, CMake, make, wget, curl, git.

Note

For detailed instructions on how to build the HPC-Stack on two particular configurations of MacOS, see Chapter 3

To determine whether these prerequisites are installed, query the environment variables (for Lmod) or the location and version of the packages (for cmake, make, wget, curl, git). For example:

echo $LMOD_PKG
which cmake
cmake --version

If compilers or MPI’s need to be installed, consult the HPC-Stack Prerequisites document for further guidance. MPI could also be built bundled with the HPC-stack, if not pre-installed in the system.

2.2.2. Ubuntu 20.04 Linux Example

The example for Ubuntu 20.4 is for a user with sudo privileges to install system-wide. First, install GNU 10 compilers:

sudo apt install gcc-10 g++-10
sudo apt install gfortran-10
which gcc

The location of default compilers is likely be in /usr/bin/ (e.g. /usr/bin/gcc), and other versions are installed with the same location with the version tag. Check all the versions installed and configure the alternatives to specify which version is to be default. Below is the example when two versions are available, e.g., gcc-9 and gcc-10.

ls -la /usr/bin/gcc-*
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-10 1
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 2
sudo update-alternatives --config gcc

Make selection (1) for gcc-10 to be the first priority (default).

Repeat similar steps to configure the alternatives for g++ and gfortran. No need to configure for gfortran if you have only a single version.

ls -la /usr/bin/g++-*
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-10 1
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-9 2
sudo update-alternatives --config g++

Install lua, luac and tcl needed to support the Lmod module management system. Verify the installation location.

sudo apt-get install lua5.3 liblua5.3-0 liblua5.3-dev lua-filesystem-dev lua-posix-dev
sudo apt-get install tcl tcl8.6-dev libtcl8.6
which lua
which luac

Standard installation paths for lua/luac are /usr/bin/lua and /usr/bin/luac. Download and install Lmod, the module management environment, if not installed in the system. When installed, the environmental variable $LMOD_ROOT is defined:

echo $LMOD_ROOT
git clone https://github.com/TACC/Lmod.git
cd Lmod

See the INSTALL file for instructions; configure and install. Use the paths for lua and luac installed in the previous step. Lmod could be configured to be installed system-wide, or in user space as in the example below, e.g., in ${HOME}/apps.

./configure --with-lua=/usr/bin/lua  --with-luac=/usr/bin/luac --prefix=${HOME}/apps
make install

Add the Lmod environment initialization to your shell profile, i.e. to $HOME/.bashrc or $HOME/.bash_profile:

export BASH_ENV=$HOME/apps/lmod/lmod/init/profile
source $BASH_ENV

Install wget, git, make and cmake; cmake could also be built bundled with the hpc-stack.

sudo apt install git
sudo apt install wget
sudo apt install make
sudo apt install cmake

Install openssl, libssl-dev packages:

sudo apt install openssl
sudo apt-get install libssl-dev

Both python2 and python3 are needed; python2 version higher than 2.7.x; python3 version higher than 3.6. You could update python3 if higher versions are available (python3.9 in the example below at the time of writing), and set the alternatives when multiple versions exist. The highest version among the existing python3.8 and python3.9 is set as a default in the example below (choose selection 1 for python3.9 when prompt).

which python3
python3 --version
apt list | grep python
sudo apt install python3.9
sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.9 1
sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.8 2
sudo update-alternatives --config python3
python3 --version

Verify the version (python 3.9) set as default after the configuration.

2.2.3. Configure the Build

Choose the COMPILER, MPI, and PYTHON version, and specify any other aspects of the build that you would like. For Level 1 systems, a default configuration can be found in the applicable config/config_<platform>.sh file. For Level 2-4 systems, selections can be made by editing the config/config_custom.sh file to reflect the appropriate compiler, MPI, and Python choices for your system. If Lmod is installed on your system, you can view package options using the module avail command.

Some of the parameter settings available are:

  • HPC_COMPILER: This defines the vendor and version of the compiler you wish to use for this build. The format is the same as what you would typically use in a module load command. For example, HPC_COMPILER=intel/2020. Use gcc -v to determine your compiler and version.

  • HPC_MPI: This is the MPI library you wish to use. The format is the same as for HPC_COMPILER. For example: HPC_MPI=impi/2020.

  • HPC_PYTHON: This is the Python interpreter to use for the build. The format is the same as for HPC_COMPILER, for example: HPC_PYTHON=python/3.7.5. Use python --version to determine the current version of Python.

Other variables include USE_SUDO, DOWNLOAD_ONLY, NOTE, PKGDIR, LOGDIR, OVERWRITE, NTHREADS, MAKE_CHECK, MAKE_VERBOSE, and VENVTYPE. For more information on their use, see HPC-Stack Parameters.

Note

If you only want to install select components of the HPC-Stack, you can edit the stack/stack_custom.yaml file to omit unwanted components. The stack/stack_custom.yaml file lists the software packages to be built along with their version, options, compiler flags, and any other package-specific options. A full listing of components is available in the HPC-Stack Components section.

2.2.4. Set Up Compiler, MPI, Python & Module System

Note

This step is required if you are using Lmod modules for managing the software stack. Lmod is installed across all Level 1 and Level 2 systems and in the containers provided. If LMod is not desired or used, the user can skip ahead to Step 2.2.5.

After preparing the system configuration in ./config/config_<platform>.sh, run the following command from the top directory:

./setup_modules.sh -p <prefix> -c <configuration>

where:

<prefix> is the directory where the software packages will be installed during the HPC-Stack build. The default value is $HOME/opt. The software installation trees will branch directly off of <prefix>, while the module files will be located in the <prefix>/modulefiles subdirectory.

Attention

Note that <prefix> requires an absolute path; it will not work with a relative path.

<configuration> points to the configuration script that you wish to use, as described in Step 2.2.3. The default configuration file is config/config_custom.sh.

Additional Options:

The compiler and MPI modules can be handled separately from the rest of the build in order to exploit site-specific installations that maximize performance. In this case, the compiler and MPI modules are preceded by an hpc- label. For example, to load the Intel compiler module and the Intel MPI (IMPI) software library, enter:

module load hpc-intel/2020
module load hpc-impi/2020

These hpc- modules are really meta-modules that load the compiler/MPI library and modify the MODULEPATH so that the user has access to the software packages that will be built in Step 2.2.5. On HPC systems, these meta-modules load the native modules provided by the system administrators.

In short, you may prefer not to load the compiler or MPI modules directly. Instead, loading the hpc- meta-modules as demonstrated above will provide everything needed to load software libraries.

It may be necessary to set few environment variables in the build_stack.sh script. For example: r

export PATH=/usr/local/sbin:/usr/local/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/lib64:/usr/local/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH

Lmod needs to be initialized based on the installation directory:

module purge
export BASH_ENV=<Lmod-installation-dir>/lmod/lmod/init/profile
source $BASH_ENV
export LMOD_SYSTEM_DEFAULT_MODULES=<module1>:<module2>:<module3>
module --initial_load --no_redirect restore
module use <$HOME>/<your-modulefiles-dir>

where:

  • <Lmod-installation-dir> is the top directory where Lmod is installed

  • <module1>, ...,<moduleN> is a comma-separated list of modules to load by default

  • <$HOME>/<your-modulefiles-dir> is the directory where additional custom modules may be built with Lmod (e.g., $HOME/modulefiles).

2.2.5. Build the HPC-Stack

Now all that remains is to build the stack:

./build_stack.sh -p <prefix> -c <configuration> -y <yaml_file> -m

Here the -m option is only required when you need to build your own modules and LMod is used for managing the software stack. It should be omitted otherwise. <prefix> and <configuration> are the same as in Step 2.2.4, namely a reference to the absolute-path installation prefix and a corresponding configuration file in the config directory. As in Step 2.2.4, if this argument is omitted, the default is to use $HOME/opt and config/config_custom.sh respectively. <yaml_file> represents a user configurable yaml file containing a list of packages that need to be built in the stack along with their versions and package options. The default value of <yaml_file> is stack/stack_custom.yaml.

Warning

Steps Step 2.2.3, Step 2.2.4, and Step 2.2.5 need to be repeated for each compiler/MPI combination that you wish to install. The new packages will be installed alongside any previously-existing packages that may already have been built from other compiler/MPI combinations.

From here, the user can continue to install and run applications that depend on the HPC-Stack.