SUMMA-Actors: Structure for Unifying Multiple Modeling Alternatives with Actors

SUMMA-Actors is a modified version of the already existing SUMMA software that can be found here. SUMMA-Actors uses the Actor Model to increase scalability and fault-tolerance. It is built using the C++ Actor Framework.

Documentation

A more in-depth documentation can be found here. SUMMA-Actors depends on many files from the original SUMMA repo. Below is a quick start guide for compiling and running SUMMA-Actors, please consult our wiki for more in-depth documentation. Consider creating an issue for any missing documentaion, asking questions, or providing suggestions.

Directory Structure

SummaActors is set up with the following sub-directoies, we will consider the top level Summa Actors directory the root_dir:

• bin
• build
  • cmake
  • includes
  • makefiles
  • source
  • summa (https://github.com/CH-Earth/summa)
• utils
• README.md

First clone Summa-Actors to your workstation. Then cd into build/ and clone summa or summa-sundials into Summa-Actor's build directory.

Compiling Summa-Actors

To compile SUMMA-Actors, use cmake with the CMakeLists.txt located in build/cmake/. You have the option to compile with or without the sundials library, definded by the line option(SUNDIALS "Use SUNDIALS" ON) in the CMakeLists.txt file. If set to ON, it compiles with sundials. If set to OFF, it compiles without sundials.

If compiling with sundials make sure to install the sundials IDA solver version 6.3.0 before attempting to compile SUMMA-Actors. Once installed specifiy the instalation directory of sundials by modifying the line set(DIR_SUNDIALS "/path/to/sundials/libs") in the CMakeLists.txt file.

Before compiling, make sure to install the following dependencies:

• g++
• gfortran
• NetCDF-Fortran
• OpenBLAS
• C++ Actor Framework

Here are the steps to compile SUMMA-Actors:

• cd into build/cmake/
• create a build directory within the build/cmake/ directory
• cd into build/cmake/build
• run cmake ..
• run make -j
• The summa_actors executable is created in the bin/ directory.

SUMMA-Actors supports four build types: Debug, Cluster, Release, and Cluster_Debug. The default build type is Release. You can set the build type by using the -DCMAKE_BUILD_TYPE=build_type option, where build_type is one of the four options listed above. To compile with the Cluster build type, make sure to load the following modules with module load before compiling when working on clusters:

• gcc/9.3.0
• netcdf-fortran
• openblas
• caf

Running SUMMA-Actors

Running SUMMA-Actors is done with the following command: ./summa_actor -g startGRU -n numGRU -c path_to_config_file

The values for -g and -n are integers where -c is the full path to the configuraiton file for summa actors.

The configuration file is a json file. The contents of the JSON file are below:

{ "Distributed_Settings": { "distributed_mode": false, "servers_list": [{"hostname": "simlab01"}, {"hostname": "simlab05"}], "port": 4444, "total_hru_count": 517315, "num_hru_per_batch": 1000 },

"Summa_Actor": {
  "max_gru_per_job": 500
},

"File_Access_Actor": {
  "num_partitions_in_output_buffer": 4,
  "num_timesteps_in_output_buffer": 500
},

"Job_Actor": {
  "file_manager_path": "/gladwell/kck540/Sundials_Settings/fileManager_actors.txt",
  "output_csv": false,
  "csv_path": ""
},

"HRU_Actor": {
  "print_output": true,
  "output_frequency": 1000
}

}

The settings above should work for most use cases, some of the feautures we want to automate such as max_gru_per_job. However, the only field that you should have to adjust is the file_manager_path. This is the path to the file that manages the complete configuration of the SUMMA simulation. The summa confiuration files are explained in more depth in the follwoing (documentation)[https://summa.readthedocs.io/en/latest/input_output/SUMMA_input/]

Credits

The inital implementation of SUMMA is credited to the inital publications below. These publications can be found in Water Resources Research.

• Clark, M. P., B. Nijssen, J. D. Lundquist, D. Kavetski, D. E. Rupp, R. A. Woods, J. E. Freer, E. D. Gutmann, A. W. Wood, L. D. Brekke, J. R. Arnold, D. J. Gochis, R. M. Rasmussen, 2015a: A unified approach for process-based hydrologic modeling: Part 1. Modeling concept. Water Resources Research, doi:10.1002/2015WR017198.

• Clark, M. P., B. Nijssen, J. D. Lundquist, D. Kavetski, D. E. Rupp, R. A. Woods, J. E. Freer, E. D. Gutmann, A. W. Wood, D. J. Gochis, R. M. Rasmussen, D. G. Tarboton, V. Mahat, G. N. Flerchinger, D. G. Marks, 2015b: A unified approach for process-based hydrologic modeling: Part 2. Model implementation and case studies. Water Resources Research, doi:10.1002/2015WR017200.

We also credit the original creators of the C++ Actor Framework which allowed us to implement the actor model into SUMMA-Actors. Links to their research work can be found below.

• Charousset, D., Schmidt, T. C., Hiesgen, R., Wählisch, M., 2013: Native actors: a scalable software platform for distributed, heterogeneous environments. AGERE!, doi:10.1145/2541329.2541336.

• Charousset, D., Schmidt, T. C., Hiesgen, R., 2016: Revisiting actor programming in C++. Computer Languages, Systems & Structures, [doi:10.1016/j.cl.2016.01.002](http:// dx.doi.org/10.1016/j.cl.2016.01.002)