A patch-based data reorganization method for coupling large-scale simulations and parallel visualization

Abstract

The scale of some datasets generated by simulations on tens of thousands of cores are gigabyte or larger per output step. It is imperative that efficient coupling of these simulations and parallel visualization. A patch-based data reorganization method was presented for this coupling through a parallel file system. Based on the method, simulation data sets in application codes are reorganized by patch and written into many files in parallel. These datasets in these files can be read directly by visualization software with low I/O overheads. For two real simulations on above 30000 cores, large-scale datasets have been generated and visualized efficiently. © 2013 Springer-Verlag Berlin Heidelberg.