| Joint Performance Evaluation and Optimization of Two CFD Codes on Commodity Clusters (2008) | |||||||||||||||
Abstract | |||||||||||||||
| The performance analysis of computational fluid dynamics code can yield unexpected results, and translating the performance data into code improvements is not always a straightforward task. A possible approach for increasing comprehension is to look at the performance of multiple codes simultaneously. In this manner, differences in the codes may be connected to differences in the performance, allowing for the better methodology to be adopted across the codes. This paper presents such a comparative analysis for two CFD codes, the structured two-dimensional code GHOST and the unstructured two-/three-dimensional code UNCLE, both in-house codes at the University of Kentucky. These codes are used regularly across several commodity cluster platforms, two of which are briefly described. This paper presents the initial performance evaluation of these codes followed by a discussion of some of the optimization techniques applied and the resulting changes in performance. Nomenclature f = generic function, interfacial flux i,j = array, spatial indices LB,LIMB = load balance, imbalance for grid decomposition m = subiteration index = average subgrid size over multiple nodes | |||||||||||||||
Publication details | |||||||||||||||
| |||||||||||||||