Parallel chemistry acceleration algorithms based on ISAT method in gaseous detonation computations

Abstract

Based on in situ adaptive tabulation (ISAT) method [11], a series of parallel chemistry acceleration algorithms for simulation of unsteady, compressible, reactive flows are proposed. These algorithms are then applied in two-dimensional gaseous detonation computations to identify the acceleration performances of chemical reactions. The algorithms are the combinations of two types of parallel strategies, one is called as primary strategy which includes Purely Local Processing (PLP) and Transposed Processing (TP), another is called as balanced strategy which contains Data Apportion Processing (DAP) and Data Exchange Processing (DEP). In according to the choice of processing methods in the primary strategy, these algorithms are divided into two groups, that is, PLP-based algorithms (single PLP, PLP/DEP, PLP/DAP, PLP/DAP1 and PLP/DAP2) and TP-based algorithms (single TP, TP/DEP and TP/DAP). For all of these algorithms, a node deletion technique of a data table in ISAT is employed to avoid the excessive growth of the table size due to the unsteady feature of chemically reactive flow computations. The computational results show that the TP-based algorithms give the better computational accuracies than the PLP-based algorithms, compared with the computational results by direct integration (DI) algorithm. For the computational efficiency, all algorithms show the chemical speedup ratios ranged from 3.00 to 4.60 except for the single PLP algorithm with a speedup ratio of 2.50, after the 10,000 time-steps. The combined algorithms between the primary and balanced strategies have larger chemical speedup ratio compared with the single-algorithms only based on the corresponding primary strategy. It is found that a balance among table operations (including query, enlargement, insertion, retrieval and deletion of nodes in a table) of ISAT at the different parallel sub-zones of computational domain has the obvious influence on the speedup ratio of algorithms. The balance can be characterized by the deviations of the table deletion frequency or the accumulated CPU time of table operation among the tables. However, the balance is not the only index of the speedup ratio of algorithms, the synchronization of the table operations has also the important influence on the computational efficiency, which still is the open issue. Lastly, a principle for strategy/algorithm selection of parallel chemistry acceleration computations for unsteady, compressible, reactive flow is presented.