A PARALLELIZATION STRATEGY FOR THE DISCRETE TRANSFER RADIATION MODEL

Abstract

An efficient strategy for the implementation of the discrete transfer radiation model on a parallel computer architecture is presented. The parallel version of the radiation model is evaluated by analyzing timing data for the computation of four test problems where the major parameters controlling numerical accuracy and computational cost are systematically varied. A semiempirical model for the parallel run time, calibrated using the timing data, is described. The model's range of application is restricted to the IBM SP2 computer architecture; however, it could be recalibrated for any parallel computer.