Volume Rendering of Scientific Data on the T3D

We discuss a direct volume rendering system for regular structured data imple- mented on the T3D. This paper discusses how to efficiently deal with a large volume dataset from an integrated scientific application, in this case, a global climate modeling system. Because of the large size volume, workstations are inherently unsuitable for storing even one volume let alone a series of time steps. Using the T3D SHMEM library, we were able to make use of asynchronous data transfer and employ communications overlap with computation. In addition, using an object space parallel decomposition of the dataset allows the algorithm to reduce cache misses on access to the volume data. The system is currently being tested and implemented on the T3D at JPL and we report on the results obtained.