Parallel Model Simplification of Very Large Polygonal Meshes

Working with polygonal models that contain tens of millions of polygons or more is difficult because such models tend not to fit into the core memory of a typical workstation. Even models of only a few million polygons are too large. A typical workstation has a gigabyte of memory, but these large models require an order of magnitude more. Rendering and other manipulations of polygonal models often requires that they be completely contained in memory. Even with large amounts of swap the processing tools will page fault, which severely increases the computation time. So, to handle large models the algorithms must be parallelized and the data distributed across a number of processors. One such class of algorithms is mesh simplification. Given R-Simp, a sequential model simplification algorithm, we derive a parallel implementation (PR-Simp). Our results show that very large models can be simplified with minimal degradation to model quality due to the partitioning of the data. Other advantages of PR-Simp include good speedup and scalability. Despite inherent difficulties of parallelizing mesh simplification algorithms without degradation of output quality, we believe that PR-Simp is a step in the right direction.