TY - GEN
T1 - Concepts for efficient flow solvers based on adaptive Cartesian grids
AU - Muntean, Ioan Lucian
AU - Mehl, Miriam
AU - Neckel, Tobias
AU - Weinzierl, Tobias
PY - 2009
Y1 - 2009
N2 - This contribution describes mathematical and algorithmic concepts that allow for a both numerically and hardware efficient implementation of a flow solver. In view of numerical efficiency, this strongly suggests multigrid solvers on adaptively refined grids in order to minimize the amount of data to be computed for a prescribed accuracy as well as the number of iterations. In view of hardware efficiency, a minimization of memory requirements and an optimization of data structures and data access tailored to the memory hierarchy of supercomputing architectures is essential, since flow solvers typically are data intensive applications. We address both the numerical and the hardware challenge with a combination of structured but flexible adaptive hierarchical Cartesian grids with space-filling curves as traversal scheme and stacks as data structures. These basic concepts are applied to the two computationally demanding application areas turbulent flow simulations and fluid-structure interactions.We show the benefits of our methods for these applications as well as first results achieved at the HLRB2 and smaller clusters.
AB - This contribution describes mathematical and algorithmic concepts that allow for a both numerically and hardware efficient implementation of a flow solver. In view of numerical efficiency, this strongly suggests multigrid solvers on adaptively refined grids in order to minimize the amount of data to be computed for a prescribed accuracy as well as the number of iterations. In view of hardware efficiency, a minimization of memory requirements and an optimization of data structures and data access tailored to the memory hierarchy of supercomputing architectures is essential, since flow solvers typically are data intensive applications. We address both the numerical and the hardware challenge with a combination of structured but flexible adaptive hierarchical Cartesian grids with space-filling curves as traversal scheme and stacks as data structures. These basic concepts are applied to the two computationally demanding application areas turbulent flow simulations and fluid-structure interactions.We show the benefits of our methods for these applications as well as first results achieved at the HLRB2 and smaller clusters.
UR - http://www.scopus.com/inward/record.url?scp=84897681498&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:84897681498
SN - 9783540691815
T3 - High Performance Computing in Science and Engineering, Garching/Munich 2007 - Transactions of the 3rd Joint HLRB and KONWIHR Status and Result Workshop
SP - 535
EP - 549
BT - High Performance Computing in Science and Engineering, Garching/Munich 2007 - Transactions of the 3rd Joint HLRB and KONWIHR Status and Result Workshop
T2 - 2007 3rd Joint HLRB and KONWIHR Result and Reviewing Workshop
Y2 - 3 December 2007 through 4 December 2007
ER -