Extended Abstract (60K)11.8
Porting the High Resolution Limited Area Forecast System (HLAFS) on shared and distributed memory hardware platforms for real-time operational weather forecasting in China
Zaphiris Christidis, IBM Thomas J. Watson Research Center, Yorktown Heights, NY; and Z. Zhu
The China Meteorological Administration (CMA) High Resolution Limited Area Forecasting System (HLAFS) has been ported from Vector architectures into the IBM RS6000 SP distributed memory parallel system. HLAFS is regional model used in production at the CMA on a 0.5 and 0.25 degrees of horizontal resolution. It employs explicit finite differencing techniques in time, and second order finite differencing schemes in space. The code was parallelized using standard domain partitioning techniques, and MPI was implemented to facilitate communication of data among different compute nodes. In order to take into advantage the shared memory architecture within each node of the SP system, the code was altered and OpenM was introduced via compiler directives. The structure of the code was significantly modified in order to execute optimally on the SP system, by taking into advantage various hardware factors such as processor connectivity and cache. In this paper we describe the different approaches and steps taken as well as the various techniques used in order to satisfy the strict operational requirements at CMA.
Session 11, European and other International Applications
Wednesday, 16 January 2002, 1:30 PM-4:59 PM
Previous paper Next paper