Title:	EVALUATION OF PROGRAMMING MODELS FOR ATMOSPHERIC APPLICATION
Author(s):	Carla Osthoff, Roberto Pinto Souto, Fabrício Vilasbôas, Pedro L.Silva Dias, Francieli Boito, Rodrigo Kassick, Claudio Schepke, Philippe Navaux, Jairo Panetta, Pedro Pais Lopes
ISBN:	978-989-8533-06-7
Editors:	Hans Weghorn, Leonardo Azevedo and Pedro Isaías
Year:	2011
Edition:	Single
Keywords:	Multi-core, GPU, Atmospheric Numerical Simulation Model, High Performance Computing
Type:	Full Paper
First Page:	275
Last Page:	282
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Ocean-Land-Atmosphere Model (OLAM) is a global numerical simulation model initially developed at Duke University, USA. As it frequently happens in many areas of science, OLAM parallelism uses domain decomposition techniques. Since this application has a high computational load, it becomes a good candidate to evaluate distributed systems, multi-core and manycores environments. In this context, this paper discuss the parallel performance of OLAM implementations in multi-core and manycore environments, under the following programming models: (1) a parallel MPI implementation running one process per core on the multi-core node, (2) a hybrid parallel MPI/OpenMP implementation running one process per node and one thread per core on the multi-core node, (3) a hybrid parallel MPI/OpenMP-Accelerator implementation running one process per node and one thread per core on the multi-core node and (4) a hybrid parallel MPI/CUDA implementation running one process per core and each process running threads on the manycore GPU node device. The results show that the adopted programming model impacts significantly the performance of the application. We show that as we increase the number of cores, the hybrid parallel MPI/OpenMP implementation running threads on each core executes faster than any other implementation.
	Go Back