Aceleración de la microfísica de lluvia para “modelo avanzado de predicción del estado del tiempo wrf” utilizando computación heterogénea paralela

Abstract

The current weather forecast use mainly numerical models to resolve the atmospheric dynamics, allow to establish future conditions based on initial conditions of atmospheric variables. Given the number of meteorological variables, complex systems of nonlinear equations and numerical methods used, it is necessary to divide the analysis region in grids of a fixed size, the size of grid determines the resolution of the model and depending on the resolution, the results the prognosis may have chance of success. Increasing the resolution means decreasing the temporal spacing of the grid points and thus increase the computing power required by the model dynamics; by increasing computational power aspects such as intercom nodes, access to local and remote memory access to data and distributing them throughout the involved solution and message passing it is involved to coordinate the work of distributed processing involved: for these reasons is that several studies have established that increasing the resolution 2 times requires increased computing power by 10 times, with the associated economic costs. In the last decade, the use of graphics and vector accelerators, like the use of FPGAs have allowed large computing power is achieved with much simpler configurations and models consistent programming, such architectures that combine processors and accelerators has been called heterogeneous computing. In the research presented different techniques used to accelerate the forecast model WRF in heterogeneous computing platforms are exposed.