Volume rendering on multiprocessor architecture with shared memory. A concurrent volume rendering algorithm

Volume rendering, i.e. direct rendering of regular volume data and not via surface primitives, is a task of scientific visualization requiring huge computing power. Modern multiprocessor graphic workstations can provide such powerful computing resources, if the rendering algorithm is suitable to be implemented concurrently. The important features of different multiprocessor architectures are discussed taking into account the effort to parallelize an existing sequential volume rendering algorithm. The tools required for concurrent programming on graphic workstations are described. This includes dynamic load balancing as well as the considerations to employ these tools efficiently. Our design and implemetation of a volume rendering algorithm using raycasting on a shared memory multiprocessor system is introduced. This algorithm avoids to be influences by the negative effect on the achievable parallel speedup on a shared memory architecture due to the memory bus contention. This includes strategies for data rearangement, data partitioning, work partitioning and dynamic load balncing. The results and performance measurements are discussed under consideration of Amdahl's Law which formulates an upper limit for the achievable speedup of a parallel program