Key points are not available for this paper at this time.
An algorithm for computing ray traced pictures is presented, which adaptively subdivides scenes into S subregions, each with roughly uniform load. It can yield speedups of O ( S 2/3 ) over the standard algorithm. This algorithm can be mapped onto a parallel architecture consisting of a three dimensional array of computers which operate autonomously. The algorithm and architecture are well matched, so that communication overhead is small with respect to the computation, for sufficiently complex scenes. This allows close to linear improvements in performance, even with thousands of computers, in addition to the improvement due to subdivision. The algorithm and architecture provide mechanisms to gracefully degrade in response to excessive load. The architecture also tolerates failures of computers without errors in the computation.
Dippé et al. (Sun,) studied this question.