SC²S Colloquium - June 25, 2014
|Date:||June 25, 2014|
|Time:||3 pm, s.t.|
Dominik Gutermuth: Application of space-filling curves on unstructured grids for simulations on a sphere
Historically, climate and weather simulations were usually computed on latitude-longitude grids. Although they provide several advantages, there are also some drawbacks, e.g. the so-called polar problem for simulations on earth-scale. Therefore, the use of alternative grids is of growing interest. In this thesis, the two most promising alternatives are used: cubed sphere and hexagonal grids.
For simulations on HPC architectures, the consideration of the memory hierarchy successively got a crucial point for performance improvements. Space-filling curves (SFCs) provide several beneficial properties such as spatial and temporal locality. We use three different SFCs to reorder the originally created grid data (i.e. faces, edges and vertices). The reordering of the grid primitives is evaluated with a serial and parallel shallow water simulation which is computed on a dual grid.
The results show, that hexagonal grids can be improved, while the straight-forward grid generation for cubed sphere grids already provides a proper order.
Florian Klein: Implementation and Evaluation of Simulations with Triangular Grids on the Sphere
Climate and weather are two typical examples which require solving partial differential equations on the sphere. In the present thesis, we focus on solving similar equations on a triangular grid, yielding several challenges, e.g. translation of the three-dimensional spherical surface to a two-dimensional grid, using a cubed sphere projection. The goal is the efficient simulation of hyperbolic equations on a global scale for which the consideration of the spherical shape of the earth is significant.
The purpose of this Master’s thesis is to revise existing work and to extend it with equidistant and equiangular projections between the regular two-dimensional grid and the three-dimensional distorted triangles on the sphere. For validation, we implemented different benchmarks including a radial dam break, a solid body rotation and a deformational flow benchmark.
Navier-Stokes equations are frequently used as a crucial component of today's fluid-simulations. With growing number of many-core systems such as GPUs, XeonPhi and SIMD units on CPUs, solvers for the Navier-Stokes equations have to be redesigned to match to such a massively parallel layout. This redesign of one particular Navier-Stokes simulation mainly used at our Chair for Scientific Computing in Computer Science (SCCS) represents the major task of this IDP as well as an interactive visualization to support understanding the simulation.