Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Computational Science, Engineering & Technology Series
TRENDS IN PARALLEL, DISTRIBUTED, GRID AND CLOUD COMPUTING FOR ENGINEERING
Edited by: P. Iványi, B.H.V. Topping
Massive Parallelization for Industrial CFD Based Analysis and Design
S. Peigin and B. Epstein
The Academic College of Tel-Aviv Yaffo, Tel-Aviv, Israel
S. Peigin, B. Epstein, "Massive Parallelization for Industrial CFD Based Analysis and Design", in P. Iványi, B.H.V. Topping, (Editors), "Trends in Parallel, Distributed, Grid and Cloud Computing for Engineering", Saxe-Coburg Publications, Stirlingshire, UK, Chapter 12, pp 269-290, 2011. doi:10.4203/csets.27.12
Keywords: computational efficiency, massive parallelization, parallel cooperative strategy, multilevel parallelization, industrial CFD based analysis and design, applied aerodynamics, Navier-Stokes computations.
This chapter is devoted to the problem of efficient parallelization for industrial computational fluid dynamics (CFD) based analysis and design tools in applied aerodynamics. Design and analysis of aerodynamic configurations is of major significance in reducing costs and thus improving competitiveness of aircraft manufacturing. This explains a industry demand for computationally efficient, robust and user-friendly software which implements aerodynamic analysis and optimization.
In fact, the improvements in computational efficiency are vital for the success of any analysis or optimization algorithms in an engineering environment. Massive parallelization is particularly advantageous for achieving this goal, since a highly scalable parallel implementation allows a dramatic reduction of the overall computation time.
To reach this goal, two new general approaches for parallel implementation of industrial CFD based tools for analysis and design were proposed by the authors: a parallel cooperative strategy and a multilevel parallelization strategy.
The idea of parallel cooperative strategy for iterative numerical algorithms consist in employing the information exchange between the members of the same parametric family. With this end of view, the information on already computed (or partially computed) solutions is distributed in the multilevel way which results in an essential speed-up of the whole parallel process. A sophisticated use of this (off-line and on-line) information may essentially reduce the number of iterations needed for convergence thus dramatically diminishing the overall time of multiple aerodynamic simulations. The results presented show that the proposed technology may result in the parallel efficiency which is formally higher than 100%. This is of course due to the information exchange, not feasible in the case of serial runs.
The second approach (multilevel parallelization strategy) was successfully applied within the framework of industrial CFD based software for aerodynamic design. This parallelization strategy efficiently makes use of the computational power supplied by multiprocessor systems and includes parallelization of the multiblock full Navier-Stokes solver, parallel CFD evaluation of objective function on multiple geometries, parallelization of the optimal search algorithm and, finally, parallel evaluation of optimal search on multiple search domains. An additional parallel level handles automatic grid generation for multiple geometries.
The method was applied to the problem of three-dimensional aerodynamic optimization with nonlinear constraints. The results demonstrated that the algorithm combines high computational efficiency with high accuracy of optimization. A significant computational time-saving (based on deep embedded parallelization) allowed the use of the method in a demanding engineering environment.
purchase the full-text of this chapter (price £20)