Progressive Refinement Radiosity on a Transputer Network

A parallel implementation of the progressive refinement algorithm for radiosity computation on a transputer network is described. Worker transputers compute the radiosity values of all patches using hemi-cubes, while rendering transputers generate images of the actual results. The scene is split up into parts of equal size and distributed among the workers to overcome the limitations of local memory. The workers are configured in a minimum pathlength network to reduce the communication overhead for loading non-local scene data and for updating patch radiosities.