PNNL

Abstract

Introduction Calculations on logically regular grids are common in many areas of numerical simulation and engineering. Increases in computer speed have led to corresponding increases in the accuracy and range of application of these calculations; however, the trend toward parallel computers has made realizing these gains in speed and problem size increasingly difficult. For large problems, the data must be apportioned between the allocated processors. Most algorithms require that in order to update an element at the edge of a subdomain residing on a particular processor, fresh data from neighboring processors is need