In a project funded by the NWS Total Water Initiative, IOOS and RPS/ASA will be architecting and deploying a modeling testbed using Amazon Web Services (AWS) . The project will create infrastructure for running models and explore challenges such as getting large datasets to and from the cloud in a cost efficient manner. A version of the Coupled Northwest Atlantic Prediction System (CNAPS) will be deployed for testing. Because the ultimate goal is to support the running of models that might link to the fully operational NWS models, the project will consider model run times, linking with other models with fixed timing requirements, and stability issues.
An early success of this approach is the joint University of Washington/NANOOS deployment of the LiveOcean circulation model on the Microsoft Azure platform. LiveOcean is a computer model simulating ocean water properties in the NE Pacific and Salish Sea. The project goal is to provide 3-7 day forecasts of aragonite saturation state and pH of waters entering shellfish growing areas on the coast. It is also used quasi-operationally to create 3-day circulation forecasts that are used as a resource when generating the Pacific Northwest Harmful Algal Bloom Bulletin. LiveOcean provides circulation maps and fields for variables including salinity, temperature, pH, oxygen, phytoplankton , etc. Outputs from LiveOcean are made available via the NANOOS Visualization System (NVS) [http://nvs.nanoos.org/Explorer].
The creation of a cloud-based model testbed will both complement and potentially replace legacy systems. It will complement the existing local HPC and NWS supercomputers by providing a “middle ground” with supercomputer level performance that does not conflict with current supercomputer uses. The platform will fill a need for a place to run quasi-operational models in support of HAB forecasts and other projects requiring the integration of multiple models. It will also satisfy the requirement for an operational modeling platform for models that are not easily integrated into the existing NWS supercomputer facilities. If the platform proves stable, expansible, and jobs run on it can meet operational time requirements, it could potentially replace the supercomputer resources for some applications,
While the economies of using the cloud can be complicated, this platform could potentially provide cost efficiencies over traditional supercomputing by making resources available, and charged for, only when they are needed. Part of the project will be to document the performance and costs of various configurations. While it is not a part of the current project, this kind of testbed could also be used for on-cloud analysis and synthesis of data. Moving analysis to the data in the cloud is seen as a cost efficient way to improve analyses while decreasing data transfer costs. We hope to expand to these kinds of explorations in the future.