ParaViz: A Spatially Decomposed Parallel Visualization Algorithm Using Hierarchical Visibility Ordering Cheng Zhang 1 , Scott Callaghan 2 , Thomas Jordan 2 , Rajiv K. Kalia 1 , Aiichiro Nakano 1 * , Priya Vashishta 1 1 Collaboratory for Advanced Computing and Simulations, Department of Computer Science, Department of Physics & Astronomy, Department of Chemical Engineering & Materials Science, University of Southern California, Los Angeles, CA 90089-0242, USA {chengz,rkalia,anakano,priyav}@usc.edu 2 Southern California Earthquake Center, University of Southern California, Los Angeles, CA 90089-0742, USA {scottcal,tjordan}@usc.edu Abstract. A scalable parallel visualization algorithm has been designed to visualize large datasets that are spatially decomposed onto processors of a massively parallel computer. The ParaViz algorithm, which is based on hybrid sort-first/sort-last parallel visualization, employs distributed visibility ordering to implement a scalable hierarchical depth buffer. A visibility rank is computed for each processor depending on its relative position from the viewpoint. After each processor rasterizes its own primitives, individually rendered sub-images are hier- archically reduced to a final image using their visibility ranks. This technique allows on-the- fly visualization of parallel simulation data without data migration. The algorithm has been tested on 8–1,024 processors for molecular-dynamics simulation data. In a weak-scaling test consisting of 64,000 spherical objects (atoms) per processor, the image integration time of the ParaViz algorithm is 40–50% less than that of the conventional global Z-buffer approach. A strong-scaling test involving 16,777,216 atoms achieves a parallel efficiency of 0.98 on 1,024 processors. Keywords: parallel visualization, distributed visibility ordering, sort-first and sort-last schemes, Z-buffer, hierarchical communication scheme, molecular dynamics simulation. * Corresponding Author. Email: anakano@usc.edu. International Journal of Computational Science 1992-6669 (Print) 1992-6677 (Online) © Global Information Publisher 2007, Vol. 1, No. 4, 407-421 A Spatially Decomposed Parallel Visualization Algorithm 407 GLOBAL INFORMATION PUBLISHER