Parallelization of a Modern CFD Incompressible Turbulent Flow Code

Introduction At this point in the beginning of the 21 Century there are already numerous reliable, inexpensive and widely-used commercial CFD codes. Fluent, STAR-CD and Flow3D are representative examples. In light of this, one might question the need for developing yet another incompressible flow code. But it is important to recognize that in many of the most successful commercial codes the underlying Navier--Stokes (N.-S.) solution procedures and the numerical analysis used to implement them are quite old, and we contend, in some cases outdated. In particular, most (but not all) commercial CFD software is based on one form or another of the SIMPLE algorithm (Patankar [1]). These basic approaches are not efficient for time-dependent simulations that are becoming increasingly more widely performed as computing power continues to improve. Furthermore, although most CFD code vendors now claim the ability of their product to employ large-eddy simulation (LES) for turbulence calculations, one essentially always finds that the manner in which LES has been implemented is far from what would be considered a true, theoretical LES, and in fact is what is sometimes termed “very large eddy simulation” (VLES), or essentially equivalently, unsteady Reynolds-averaged Navier--Stokes (URANS) modeling. Finally, and in many respects the most important point with regard to the present paper, is the fact that parallelization of these commercial codes has in almost all cases been done long after the codes were originally constructed; i.e., they were not designed to be parallelized at the time they were initially coded. We contend that despite the clear success of many commercial CFD codes, the shortcomings noted above are of such significance to warrant development of a new code-maybe even a new generation of codes-at least for research purposes, but probably even for commercial applications in the longer term. The work to be reported here has been underway for much of the past decade, and various portions of it have previously been reported (several times in Parallel CFD Proceedings volumes). But it has only been in the past year that this work has reached the maturity needed for complete implementation of a new code. In the remainder of this abstract (and in the full-length paper) we will briefly recall the basic system of equations that must be solved, the numerical algorithms used to accomplish this, and parallelization results for several different cases.