Future trend of parallel computing in computational sciences

Vast amount of computation is needed in large-scale scientific computing, especially fluid dynamics, electromagnetic fields, structural analysis, heat flow, climate modeling, multiparticle system, etc. In spite of the variety of such large-scale simulations, they have striking similarity on the level of numerical methods, i.e., most of the problems are formulated in terms of partial differential equations. Systems expressed and modeled as a partial differential equation have two remarkable characteristics: proximity and autonomity. High performance parallel processing should exploit those inherent parallelism.