For many business applications there is significant sensitivity to local weather conditions, especially with regard to planning, operations and decision making.  The availability of highly localized weather model-based predictions has demonstrated the ability to improve the response to, and mitigate the impact of environmental conditions on businesses and municipalities.  To enable the creation of integrated solutions focused on the forecasts of such weather impacts, IBM Research has an on-going project, dubbed “Deep Thunder”.

A key part of the Deep Thunder project is the development, implementation and operation of high performance computing (HPC) systems which provide the information technology infrastructure to enable timely generation and delivery of high-resolution numerical weather prediction forecast products coupled with customized visualizations and business analytics. The current systems support the development and production of forecast products based upon the WRF-ARW community model driving visualization and analytics with throughput sufficient for operational use. These systems run on a variety of computing platforms and technologies focused on providing integrated, optimized customer solutions updated several times per day, with forecast periods of one to three days. We will discuss some of the current, as well as future HPC systems, architectural and operational aspects of the Deep Thunder infrastructure along with rationale for our choices.