Using Containers to Help Foster Collaboration for Model Development and Testing of the Finite-Volume Cube Sphere Dynamical Core (FV3) Model

Creating a community for model development is not without its challenges. These challenges include compiling and building a model across a variety of platforms, maintaining the different dependencies, libraries, and compiler versions. One way to help alleviate this problem is through the use of software containers where one can encapsulate all the tools necessary for model development. In other areas of software development, we have seen the significant growth of containerization of applications, services, and other projects. By using containers, one can provide a common environment for model testing and development.

At the NOAA Earth System Research Laboratory, we have containerized a model development environment with the tools necessary to compile and run the FV3 model. This development serves two purposes: to provide a tool set for model development where interested individuals can modify and recompile the model to evaluate new science innovations or techniques, and to provide a way to run the FV3 model across a variety of platforms from personal systems to commercial cloud services. The container provides the necessary tools, including mpi, to run the FV3 on multi-node multi core machines. The latter capability has been successfully demonstrated through both an internal multi node cluster as well as through a cluster of compute nodes through Amazon Web Services (AWS).

This presentation will provide an overview of our research efforts into the development of containers for the FV3 model. It will include a discussion on how these containers are used with commercial cloud services, the benefits/trade-offs of using containers, and a look at where this effort is heading for the future.