Full-frame merging for sort-las polygon rendering on a multicomputer

We propose a refinement of the Sort-Last algorithms classification based on the scheduling of the rendering and merging steps. Two algorithms whose rendering and merging steps run consecutively are described. Two different approaches were taken to implement the merging step: the distributed Framebuffer approach and the Pipeline Composition approach. The load balancing problem is also discussed: a dynamic request-based mechanism is implemented at the end of the rendering phase. Another solution, the ScanlineFlow Rasterization algorithm, is described. Its main characteristic resides on the fact that both steps, rendering and merging, run concurrently. This solution has provided good results and is a viable alternative to implement sort-last algorithms on a multicomputer. The three algorithms made use of the full-frame merging technique because merging a full frame from each node is very regular and easy to implement. Our developing platform consisted of a Parsytec MultiCluster machine with sixteen processors running the Helios Operating System and using the CDL (Component Distribution Language) parallel programming language