John Dennis, CISL CSS

The advent of petascale computational resources offers the potential to resolve many important Earth System processes that have previously only been parameterized.  However, due to the architectural trend in computing systems that have replaced increases in processor frequency with steady increases in the number of processor cores, capitalizing on this potential is non-trivial.  In this talk, I will describe several innovations that enable the use of very-large-scale cyberinfrastructure for geophysical simulations. Specifically, these innovations are: a space-filling based partitioning algorithm that has enabled multiple Community Earth System Model (CESM) components to scale to very large processor counts; a parallel I/O library (PIO) that greatly reduces the memory footprint and increases the sustained I/O bandwidth for geoscience applications; and a design for a Krylov solver used for seismic tomography that significantly increases its scalability

THIS LECTURE WILL BE WEBCAST AND RECORDED

http://www.fin.ucar.edu/it/mms/ml-live.htm