Ray coherence in parallel volume rendering

Ray coherence, meaning all processing along each ray is local to a single machine, is achieved in our parallel volume rendering environment by using a workload distribution scheme that divides the image space. This allows one to avoid the compositing stage when performing standard volume rendering in a parallel rendering pipeline. More importantly, there are a number of existing algorithms for volume rendering that either benefit from or require ray coherence when being adapted to a parallel environment. We discuss several of these algorithms and adapt and implement two of them, our own improved visibility culling technique to speed up rendering when occlusion occurs and a volumetric shadowing technique that produces more realistic and informative images. We also present novel algorithms for providing a consistent load balancing and efficiently loading and rendering pieces of a subdivided data set, addressing two of the major issues for data scalable image space distributions.;Keywords: distributed computing; sort first parallelization; volume rendering; visualization; load balancing; occlusion; shadow; ray coherence. Subject terms: computer graphics; parallel processing (electronic computers); parallel algorithms; three-dimensional imaging; high performance computing; visualization data processing.