13B.4 Collaborative research and education with numerical weather prediction in a common environment enabled by software containers

Thursday, 14 January 2016: 4:15 PM
Room 355 ( New Orleans Ernest N. Morial Convention Center)
Joshua Hacker, NCAR, Boulder, CO; and J. Exby, N. Chartier, D. Gill, I. Jimenez, C. Maltzahn, and G. Mullendore

The rapid emergence of software container technology like Docker offers immediate potential to lower barriers in numerical weather prediction (NWP) research and education. A Docker container can contain one or more software applications, all necessary dependencies, and provide isolation from other containers and other installed software unless explicitly overridden. Modularity is inherent, and containers can be linked together arbitrarily to construct complex workflows. The same containers can be deployed on Windows, OS X, or Linux based machines; deployment on commercial cloud services is an easy step. Applications within containers give predictable output regardless of operating system and chip set. In many instances containers can obviate the need for difficult and time-consuming compilation efforts and redundant flow control and analysis software. It also minimizes uncertainty in results from running simulations on multiple platforms.

We present a set of Docker containers that provide both an educational and collaborative NWP environment. The widely used community Weather Research and Forecasting (WRF) model anchors the set presented here. Linked containers include software to initialize the model, run the model, create graphics from the results, and serve output to collaborators. The containers have been engineered so that multiple independent NWP instances can take advantage of available computing resources; for example a multi-model or multi-physics ensemble can be easily produced. Testing on multiple platforms, including both personal computers and commercial cloud resources, demonstrates the following: (1) how the often-difficult exercise in compiling the WRF and its many dependencies is eliminated to accelerate classroom learning and graduate research; (2) that model output obtained on different computing systems is predictable; (3) how sharing containers provides reproducible environments for conducting research.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner