Distributing data and control for ray tracing in parallel

摘要

We first briefly describe the methodology of programming ray-tracing algorithms on distributed-memory parallel computers, or DMPCs, and review previous efforts to overcome the problems of data distribution and load balancing. Then we present two algorithms designed for DMPCs and implemented on an Intel iPSC/2. We also compare the results of our experiments with them. The first algorithm, a data-oriented parallel implementation based on message passing, demonstrates how complex designing a parallel ray-tracing algorithm can be. The second algorithm shows how we can eliminate some complexity using a control-oriented parallel approach and a shared virtual memory.