A Report on the Deployment of Numerical Weather Prediction Services in the Amazon Cloud

Date: -
Location: VIRTUAL
Don Morton, Boreal Scientific Computing, Fairbanks, AK

This work is motivated by a years-long desire to deploy custom, on-demand numerical weather prediction models often under urgent conditions. Applications may range from scenarios in which a large volcanic eruption necessitates the generation of forecast wind field ensembles to drive atmospheric transport models for initial situational awareness, to the researcher needing to generate a number of simulations to meet an upcoming conference deadline. These activities tend to require a short-term “burst” of access to computational resources, and then it may be months or years before the need arises again. Yet, when the need arises it is urgent and launching of the simulations needs to be somewhat painless and straightforward.

To support flexibility in workflows, the implementation is based on an attempt to use “the Unix way” paradigm of loosely-coupled components (or, NWP services) that can be arranged in a number of configurations for custom scenarios. In this case, however, the services are often long-running with asynchronous behavior, in a cloud environment where any number of things may go wrong.

This seminar describes the work pursued as part of DTC’s visitor program, beginning with the vision of the architecture and culminating in the successful deployment and use of on-demand WRF services running in the Amazon Cloud. The primary theme that comes to mind in describing the work is, “The Good, The Bad, And The Ugly.” The goal of exploring and developing a prototype was accomplished, and many insights for future work in this area were realized.

Please email Jenny at jbolton@ucar.edu for the meeting link