Keywords
class, gcm, parallel computing
Start Date
1-7-2004 12:00 AM
Abstract
Parallel computing is a very useful tool for computing intensive and time constrained real timeproblems. Depending on the size of the grid and processors available in the cluster, a group of nodes or processesin the grid can be represented by an individual processor and it can be responsible for their computational needs.This increases the accuracy of the model by allowing finer grid sizes, also leading to savings in time. Our study,utilizes the Canadian Land Surface Scheme (CLASS), a well-tested serial general land/atmosphere interactionmodel. CLASS is a vertical one-dimensional model and spatially adjacent nodes in the grid do not interact. Thismodel computes heat and moisture fluxes for bare ground (fractional coverage by ground), ground covered withsnow (fractional coverage by snow), ground with canopy (fractional coverage by ground), and ground with bothsnow and canopy. Within each spatial grid cell, these fractions are combined. In this paper, we demonstrate theneed of parallelizing the serial CLASS model and discuss the designs to implement it. This will enable finer gridsizes leading to higher accuracy of the model and a corresponding decrease in individual processor computingtime, when compared to the serial CLASS model. It was observed that a serial farm kind of design suits ourdesign constraints and has been successfully implemented.
Functional Parallelization of a Land Surface Model in Regional Climate Modeling
Parallel computing is a very useful tool for computing intensive and time constrained real timeproblems. Depending on the size of the grid and processors available in the cluster, a group of nodes or processesin the grid can be represented by an individual processor and it can be responsible for their computational needs.This increases the accuracy of the model by allowing finer grid sizes, also leading to savings in time. Our study,utilizes the Canadian Land Surface Scheme (CLASS), a well-tested serial general land/atmosphere interactionmodel. CLASS is a vertical one-dimensional model and spatially adjacent nodes in the grid do not interact. Thismodel computes heat and moisture fluxes for bare ground (fractional coverage by ground), ground covered withsnow (fractional coverage by snow), ground with canopy (fractional coverage by ground), and ground with bothsnow and canopy. Within each spatial grid cell, these fractions are combined. In this paper, we demonstrate theneed of parallelizing the serial CLASS model and discuss the designs to implement it. This will enable finer gridsizes leading to higher accuracy of the model and a corresponding decrease in individual processor computingtime, when compared to the serial CLASS model. It was observed that a serial farm kind of design suits ourdesign constraints and has been successfully implemented.
