A LA-COMA Implementation of Parallel Volume Rendering

Object dataflow is a popular approach often used for parallel rendering. The scene is statically distributed among processors and objects are fetched and cached only on demand. Most previous object dataflow methods were implemented on shared memory architectures, and exploit image/object space coherency to reduce cache misses. In this paper, we propose an efficient object dataflow incremental rotation system on distributed memory machines. It uses a distributed-directory scheme to trace the location of objects at other nodes. The objects migrate and replicate at different processors as in Cache-Only Memory Architectures (COMA). During the animation process, the processors predict and prefetch the data that will be needed for subsequent frame generations, thus employing a look-ahead (LA) data acquisition to hide the latency of communication. Load balancing, minimizing network congestion, and optimal algorithm embedding are some of the other issues considered in the design process. The results on the Cray T3D show good load balancing and significant