Piggybacking for More Efficient Parallel Out-of-Core Isosurfacing

Abstract

A scheme for improving the efficiency of parallel isosurfacing for very large datasets is presented. The scheme is aimed at improving performance in multi-processor environments, especially for environments in which interprocessor communication limitations become a bottleneck, such as when the number of processors can scale up without commensurate scale up in inter-processor communication bandwidth. The scheme enables load-balanced computation while also limiting unnecessary communication between processors through the use of communication piggybacking and interleaving. Empirical results are also presented and suggest that the scheme reduces communication by about 15% and overall isosurfacing time by about 13% over a highly efficient non-piggybacked parallel isosurfacing approach.