Michalakes, John, National Renewable Energy
Laboratory, and Andrew Porter, Daresbury Laboratory, UK
This presentation will describe efforts to improve WRF performance
and scaling by exploiting parallelism at the node level and in high-performance
scalable I/O systems. At the node level, progress adapting WRF to use
many and multi-core processors is discussed, including the new Many Integrated
Core (MIC) architecture (subject to non-disclosure restrictions). Also, I/O
improvements developed at Daresbury Laboratory, U.K. that combine
parallel NetCDF with WRF "quilting" are described with results and
usage information.