Investigation of Strategies for the Parallel Implementation of ISAT in LES/FDF/ISAT Computations
Liuyan Lu1, Zhuyin Ren1, S. R. Lantz2, Venkatramanan Raman3, Stephen B. Pope4, and Heinz Pitsch5. (1) Cornell University, Student, Upson 135, Cornell University, ITHACA, NY 14850, (2) Cornell Theory Center, Senior Research Associate, Cornell Theory Center, (3) Center for Turbulence Research, Research Associate, 500W Stanford University, Stanford, CA 94305, (4) Cornell University, Sibley College Professor, Sibley School of Mechanical and Aerospace Engineering, Ithaca, NY 14853, (5) Center for Turbulence Research, Professor, 500W Stanford University, Stanford, CA 94305
The LES/FDF approach for turbulent combustion offers the benefits of both large eddy simulation (LES) to treat the turbulent flow, and the PDF approach to treat turbulence-chemistry interactions (in terms of the filtered density function, FDF). The approach is implemented as a particle mesh method and computationally the most expensive aspect is determining the change in particle composition over a time step due to reaction. This cost can be significantly reduced by using in situ adaptive tabulation (ISAT). In this work we investigate the computational performance of several strategies for the parallel implementation of ISAT in LES/FDF calculations. The capability of performing LES/FDF/ISAT computations of turbulent flames is developed by incorporating the ISAT algorithm in the Stanford structured large eddy simulation (LES) and composition “filtered density function” (FDF) code. The LES/FDF/ISAT simulation of a spatially developing mixing layer is used as the test case to study the performance and load balancing of different ISAT strategies for idealized turbulent flames of both hydrogen and methane. The 9-species and 35-species detailed mechanisms are employed for the hydrogen flame and the methane flame, respectively. The results show that when it is almost always possible to retrieve from the ISAT table, then using (serial) ISAT without any message passing is optimal. But when a significant number of direct integrations of the chemical kinetic equation is required, then parallel strategies are advantageous.