Parallel Implementation of Stochastic Iteration Algorithms

Abstract

The paper examines the parallel implementation of iteration type global illumination al-gorithms. The steps of iteration depend on each other, thus their parallel implementation is not as straightforward as for random walks. This paper solves the interdependency problem by applying stochastic iteration. In this framework two fundamental questions are investigated: how many processors can be efficiently used in an algorithm and how often the processors should exchange their information. These questions are answered by a theoretical model and also by simulations.