Building a Weather Research and Forecast Service on Windows Azure

Using off-the-shelf web-based Bing map service and Microsoft's public cloud infrastructure, a proof of concept WRF as a service has been created and operated for the last 2 years hosting 1000s of complete end-to-end weather forecasts around the world. As public cloud becomes more mature, this paper explores the practicality and lessons learned for running large scale weather simulation using both hybrid on-premise compute burst model and hosting the entire compute infrastructure in the public cloud. Topics covered include usability, performance, scalability, operability, and cost of running the live cloud service.