Parallel volume rendering and data coherence on the Fujitsu AP1000

Abstract

Many scientific and engineering disciplines, through physical measurements or computational simulations, generate large scale three-dimensional data sets. Both the physical size and the computational resources needed to render these data sets present a challenge to current rendering architectures and techniques. The Fujitsu AP1000 has the memory capacity and the processing speed to render large three-dimensional data sets at interactive or near-interactive speeds. A parallel version of a volume renderer has been implemented using a ray-casting technique on this architecture. The two key issues in implementing this technique on a distributed memory, MIMD machine such as the AP1000 are the work and data distribution. To perform the data distribution, a distributed virtual memory for volume data is used. The importance of utilizing the data coherence that is inherent in volume data is demonstrated through the analysis of several case studies.