César Gómez-Martín, José-Luis González-Sánchez, Javier Corral-García, Ángel Bejarano-Borrega, Javier Lázaro-Jareño: «Shared Memory Supercomputing as Technique for Computational Electromagnetics«. VII Encuentro Ibérico de Electromagnetismo Computacional. 29-33. Monfragüe, Cáceres. Mayo de 2010.
Abstract
Nowadays, it is not possible to innovate and investigate computational electromagnetics without being able to process and store huge amounts of data. Supercomputers along with high-performance computing techniques are aimed to provide methodologies and tools for computational electromagnetics researchers that will help them to solve problems in a more efficient and easy way. In order to get the best of those supercomputers it is important to know which kind of computer architecture is going to be used and how to achieve the best performance applying the appropriate programming model. Shared memory supercomputing is probably the easiest
way of exploiting parallelism in computational electromagnetics, the combination of shared memory programming techniques like OpenMP with other distributed programming techniques will allow the improvement of supercomputer performance. Is is also
very important to track and test new parallel programming models such as CUDA, Intel Ct or OpenCL because they are meant to exploit modern supercomputers efficiently. Automatic parallelization models are improving performance quickly but there is not a definitive model, they all have pros and cons and whether to use one programming model or another should be thoroughly studied.
Index Terms—HPC, Supercomputing, Electromagnetism, CenitS, LUSITANIA