Object dataflow is a popular approach used in parallel rendering.The data representing the 3D scene is statically distributed amongprocessors and objects are fetched and cached only on demand. Mostprevious object dataflow methods were implemented on shared memoryarchitectures and exploited spatial coherency to reduce hardware cachemisses. We propose an efficient model for object dataflow parallelvolume rendering on message passing machines. The algorithm isintroduced and its ray storage mechanism is used to support latencyhiding by postponing computation on inactive rays. Memory usage isoptimized by letting objects migrate and replicate at differentprocessors rather than the common static assignments. Our cache onlymemory approach uses a distributed directory scheme to trace thelocation of objects at other nodes. A mechanism to minimize networkcongestion was implemented which optimizes channel utilization. Unlikeprevious methods, our approach can benefit from temporal coherence andeffectively minimizes communication costs during animation on limitedbandwidth multiprocessing environments. We report results of thealgorithm's implementation on several platforms like Cray T3D, ConvexSPP and DEC alpha cluster of workstations (COWs), and achieved higherefficiency and scalability than existing algorithms
展开▼
