MPI versus MPI+OpenMP on IBM SP for the NAS benchmarks
Proceedings of the 2000 ACM/IEEE conference on Supercomputing
Parallel computation for spherical harmonic synthesis and analysis
Computers & Geosciences
Parallel Programming: for Multicore and Cluster Systems
Parallel Programming: for Multicore and Cluster Systems
Hi-index | 0.00 |
This paper describes a hybrid MPI/OpenMP approach to using diskless computing clusters to solve time-consuming problems related to ultra-high-degree and -order spherical harmonic analysis using the EGM2008 geopotential model with least-squares collocation to improve the accuracy of EGM2008-derived geoid heights. The proposed hybrid approach conducted 2190-degree and -order spherical harmonic analysis and least-squares collocation fitting using a cluster system of low-performance nodes and achieved a maximum parallel efficiency of 99%. Calculation of the associated Legendre functions and numerical stability were ensured for both 64 and 128bits in an EGM2008 spherical harmonic analysis for the Korean Peninsula using parallelization modeling. EGM2008 not only described the gravity field of the Korean Peninsula well, but its results were also very accurate compared to those of other global geopotential models. EGM2008-derived geoid heights were fitted with GPS/leveling data from least-squares collocation, which improved their accuracy. For this procedure, parallelization modeling was used to explore the optimum correlation length representing the highest accuracy. After the geoid heights were fitted, their accuracy was evaluated for several cases to determine the optimum distance of GPS/leveling points for least-squares fitting around the Korean Peninsula, and to determine the optimum correlation length by comparing the various results. A correlation length of 119.4km showed the best fitting results for the case using 102 GPS/leveling points with 40-km intervals.