Cornell Box [ZWHS95]
Cornell University Program of Computer Graphics
Cornell Seal

Efficient parallel global illumination using density estimation.

David Zareski, Bretton Wade, Philip Hubbard, and Peter Shirley.

In Proceedings of Visualization '95 - Parallel Rendering Symposium, pages 219--230, October 1995.

This paper presents a multi-computer, parallel version of the recently-proposed ``Density Estimation'' (DE) global illumination method, designed for computing solutions of environments with high geometric complexity (as many as hundreds of thousands of initial surfaces). In addition to the diffuse inter-reflections commonly handled by conventional radiosity methods, this new method can also handle energy transport involving arbitrary non-diffuse surfaces. Output can either be Gouraud-shaded elements for interactive walkthroughs, or ray-traced images for higher quality still frames. The key difference of the DE algorithm from conventional radiosity, in terms of its ability to parallelize efficiently, is its microscopic view of energy transport, which avoids the O(n*n) pairwise surface interactions of most previous macroscopic radiosity algorithms (i.e., those without clustering). Parallel DE is implemented as two separate parallel programs which perform different phases of the DE method. The first program performs the particle-tracing phase, and the second performs the density-estimation and meshing phases. Each parallel program consists of a single master task and multiple worker tasks executing on separate workstations connected over a local area network. Communication is performed using the PVM software package and a shared file system. The goal of this effort is to provide a near-linear speedup for solutions to existing environment models using tens of processors. The parallel efficiency of the first program has been measured to be above 90% for as many as 16 workers, and the parallel efficiency of the second program has been measured to be above 70% for as many as 12 workers.

This paper is available as a compressed Postscript file ( 70K).

Last updated 08/29/02 PCG www Home