The subzone load-on-demand (SZL) visualization architecture has been presented in prior work. Subsequent work has extended it to support in-situ visualization for steady-state CFD simulations. In this work, we further extend it to support in-situ visualization of unsteady CFD simulations. We analyze its performance using four unsteady data sets: an unsteady multi-block structured CFD simulation of a wind turbine, an unsteady unstructured-grid hurricane simulation, and synthetic structured and unstructured data. We then examine performance scaling and project the utility of this approach in view of the anticipated growth of CFD simulations in the next decade. Compared with full data set output, the data reduction due to in-situ extraction ranges from 83% for the hurricane simulation to 99.5% for the largest structured synthetic data. This advantage is shown to increase as data set size increases. The data size required for unsteady iso-surfaces are shown to scale with O(n2/3), where n is the number of cells in the grid, making this approach a viable candidate for the trillion-cell simulations expected in the next decade.
展开▼
