Adaptation of volume visualization techniques to MIMD architectures using MPI

This paper presents a divide-and-conquer approach to developing visualization software for scientific analysis of high resolution, volume datasets on, distributed memory parallel computers. A direct volume rendering method is discussed and the design of the parallel implementation of the original sequential algorithm is highlighted. For this algorithm, the data, once distributed, remains in place regardless of the viewer's location, and no interprocessor communication is required during scene generation. The software was developed using an early version of the Message Passing Interface (MPI), and the MPI features used in this application are discussed. Finally, test results indicating the performance of our parallel implementation on an nCUBE 2 are presented and analyzed.<
>