Performance analysis of a CFD-code on the IBM-SP2
نویسنده
چکیده
1 Abstract The results presented in this report treats the performance of the the AVBP (ver.2.0) CFD-code on the IBM-SP2 parallel supercomputer. AVBP is a cell-vertex nite volume code designed to solve the partial diierential equations that govern the motion of compressible uid ow, e.g. the Navier-Stokes equations or the simpliied set of equations known as the Euler equations. AVBP is able to handle structured and unstructured meshes in both 2D and 3D. As test cases we have used the three dimensional geometries known as the ONERA M6 wing (simple wing geometry) and the Dassault Falcon aircraft (full aircraft geometry). Both these grids are unstructured with tetrahedral elements. The report will emphasize the importance of choosing an appropriate grouping of the elements since non-appropriate chosen the computation time might be more than 50 % longer than necessary. On the other hand, the report shows that the choice between the RCB (Recursive Coordinate Bisection) and the RIB (Recursive Inertia Bisection) partitioning algorithms is of less importance when one compares the computing times. 2 Introduction The eld of Computational Fluid Dynamics (CFD), which treats the motion of uid ows numerically, has become increasingly important during the last ten years. This is much due to the increase of computing power ooered by modern supercomputers. Actually, CFD is one of the big drivers on the development of advanced supercomputing technology. Re-versely, without the progress in hardware technology, CFD would not become such an expanding eld of research as it is today. Another hallmark making CFD such a wide eld of research is the fact that CFD is interdisciplinary, ranging over elds like aeronautical, chemical, civil, mechanical and even so electrical engineering. It is not diicult to understand that the engineering problems within these areas supply a rich eld of applications where CFD can be brought to use. Thus, a large interest group can easily be detected whenever a new CFD achievement is done. Most of these CFD applications are based on the resolution of the well-known Navier-Stokes equations or its no less known simpliied variant called the Euler equations. These equations describe the continuum behavior of the uid. Solving these equations in three dimensional space around an object of interest is a 1
منابع مشابه
Programming Research Group THE THEORY, PRACTICE, AND A TOOL FOR BSP PERFORMANCE PREDICTION APPLIED TO A CFD APPLICATION
The Bulk Synchronous Parallel (BSP) model provides a theoretical framework to accurately predict the execution time of parallel programs. In this paper we describe a BSP programming library that has been developed, and contrast two approaches to analysing performance: (1) a pencil and paper method with a theoretical cost model; (2) a pro-ling tool that analyses trace information generated durin...
متن کاملTheory , practice , and a tool for BSPperformance predictionJonathan
The Bulk Synchronous Parallel (BSP) model provides a theoretical framework to accurately predict the execution time of parallel programs. In this paper we describe a BSP programming library that has been developed and contrast two approaches to analysing performance: (1) a pencil and paper method; (2) a prooling tool that analyses trace information generated during program execution. These appr...
متن کاملImplementation and Performance of a Grand Challenge 3d Quasi-Geostrophic Multi-Grid code on the Cray T3D and IBM SP2
We have taken our existing auto-tasked vector Cray C-90 3d Quasi-Geostrophic Multi-Grid (QGMG) code and implemented it in a portable way on most of todays MPPs. Here we report on its performance for the Cray T3D and IBM SP2. On all 16 processors of the C-90 the code achieved 6 GGops; currently on 256 processors of the T3D we obtain almost 4 GGops and on 256 processors of the SP2 we obtain 5 GGo...
متن کاملParallel FEM Software for CFD Problems
The present paper describes the development and the performance of parallel FEM software for solving various CFD problems. Domain decomposition strategy and parallel iterative GMRES solver have been adapted to the universal space-time FEM code FEMTOOL, which allows implementation of any partial differential equation with minor expenses. The developed data structures, the static load balancing a...
متن کاملModeling Computation and Communication Performance of Parallel Scientific Applications: A Case Study of the IBM SP2
A methodology for performance analysis of Massively Parallel Processors (MPPs) is presented. The IBM SP2 and some key routines of a finite element method application (FEMC) are used as a case study. A hierarchy of lower bounds on run time is developed for the POWER2 processor, using the MACS methodology developed in earlier work for uniprocessors and vector processors. Significantly, this hiera...
متن کامل