
However, when the JPPF is used in a cluster alone, a simply master-worker parallel model is obtained. (2012) take advantage of this tool to avoid any modification of MODFLOW and related programs. This tool is very powerful and can be used as a GRID middle-ware ( Foster et al., 2001) to distribute tasks across several computing systems. (2012) describe a parallelization strategy for stochastic modeling of groundwater systems using the Java Parallel Processing Framework (JPPF). Some previous works have focused in this direction, for example Dong et al. Nowadays there exist many parallel computing platforms that can be used to alleviate this problem. The large number of realizations required by MCS can be very demanding in computing resources and the computational time can be excessive. For example, the pioneering work on stochastic hydrogeology by Freeze (1975) applies this method. Monte Carlo simulation (MCS) is an alternative for solving these stochastic models, it is based on the idea of approximating the solution of stochastic processes using a large number of equally likely realizations. An important part of it consists of solving stochastic models (stochastic partial differential equations) describing those processes in order to estimate the joint probability density function of the parameters (e.g., transmissivity, storativity) and/or state variables (e.g., groundwater levels, concentrations) of those equations or more commonly some of their moments. Stochastic hydrogeology is a field that deals with stochastic methods to describe and analyze groundwater processes ( Renard, 2007). Observamos una mejora ligera del rendimiento a medida que aumenta el número de realizaciones. Eficiencias de 0,70, 0,76 y 0,75 se obtuvieron para 64, 64 y 96 procesadores, respectivamente. Presentamos los resultados de aceleración y eficiencia para 1000, 2000 y 4000 realizaciones para diferente número de procesadores. Esta estrategia se aplicó al estudio de un acuífero simplificado en un dominio rectangular de una sola capa. Nuestro enfoque consiste en calcular las entradas iniciales para cada realización y correr grupos de estas realizaciones en procesadores separados y después calcular el vector medio y la matriz de covarianza de las mismas. Desarrollamos un script en Python usando mpi4py, a fin de ejecutar GWMC y programas relacionados en paralelo aplicando la biblioteca MPI. En este artículo se presentan los resultados de una estrategia de paralelización para reducir el tiempo de ejecución al aplicar la simulación Monte Carlo con un gran número de realizaciones obtenidas utilizando un modelo de flujo y transporte de agua subterránea.
