Generating Simplified Meshes from Distributed Volume Data

The paper specifies the design, implementation, and test of a new parallel algorithm that interleaves Marching Cubes and a variant of vertex clustering without generating big intermediary meshes, contrary to other approaches. A userprovided clustering level defines the size of the clusters, and thus, the degree of reduction. The challenge was to enable distributed and parallel mesh generation and simplification whilst avoiding explicit data exchange messages, which would slow down mesh generation because of additional synchronisation steps between the processors. This was achieved by way of using different clustering strategies for the inner and for the outer part of a volume data partition. We conclude with an outlook on the software that will be part of our network-distributed visualization system DSVR-2, to support the investigation of atmospheric convections in meteorology and climatology.