Log on:
Powered by Elgg

Javier Delgado

Javier Delgado
Friday 28th March 2008, 7:10pm
PIRE Proposal Supervisor: Dr. S. Masoud Sadjadi PIRE Partner: Barcelona Supercomputing Center Project Title: Grid Enablement of Weather Research and Forecasting Software This summer, I will be one of the first PIRE students. I will be going to Spain to work with our existing collaborators at the European Center for the Parallelization of Barcelona (CEPBA). My main application of interest for this project is the grid enablement of hurricane damage prediction software. While the means (i.e. software and many computational resources) exist for prediction of the effects of hurricanes, it is currently necessary to sacrifice resolution in order to obtain a forecast within a reasonable time period. To address this problem, it is necessary to use many computational resources. Furthermore, due to the nature of the weather forecasting software being used (WRF), making it suitable for a computational grid is not a trivial task: an intelligent scheme for distributing the computational workload must be devised. Here at FIU, we have been working on the grid-enablement of WRF using a statistical approach that learns the behavior of an application on a set of platforms and runtime configurations (such as number-of-compute-nodes, processing power, etc.) and is able to extrapolate to predict runtime on a target configuration. We have also done extensive study of the software itself. By combining our knowledge of the software and the parallelization tools developed at CEPBA, we will be able to make progress in this important research area. As a result, hurricane damage predictions of higher accuracy could be obtained, hence less personal and financial damage will be experienced in future hurricanes. We have been working with several of professors and with a graduate student from CEPBA for months now. We have learned how to use many of their tools and are in the process of publishing a paper in which we combine their tools with our own techniques of parallelization and grid enablement of applications. Both our work and theirs have resulted in multiple publications, so community acceptance has been established. The tools used at CEPBA, such as Paraver and Dimemas, are widely used in the parallel processing community. A general road map for this project is as follows. Our paper will be submitted for review shortly before we leave to Spain. When we get there, we will be able to learn more about what they have learned already using their own tools to examine WRF. By dealing with them directly, they can help us to better use their programs to realize the optimal way of making the code parallel. Once this is done, we can implement code to distribute the tasks in an optimal manner and re-run our initial tests to see how the application performs. If the performance is significantly improved, we can publish the work. Since this project deals with both the grid enablement of hurricane mitigation software and development of tools for simplifying this task, it falls into both the Cyberinfrastructure Applications and Integration Layers.