Parallel Vector Machine Research Articles

Multiobjective genetic algorithm applied to aerodynamic design of cascade airfoils

A multiobjective genetic algorithm (GA) based on Fonseca-Fleming's Pareto-based ranking and fitness-sharing techniques has been applied to aerodynamic shape optimization of cascade airfoil design. Airfoil performance is evaluated by a Navier-Stokes code. Evaluation of GA population is parallelized on the Numerical Wind Tunnel, a parallel vector machine. The present multiobjective design seeks high pressure rise, high flow turning angle, and low total pressure loss at a low Mach number. Pareto solutions that perform better than existing control diffusion airfoils were obtained.

Euler/Navier-Stokes Optimization of Supersonic Wing Design Based on Evolutionary Algorithm

This paper presents aerodynamic shape optimization of a supersonic wing for supersonic civil transportation (SST) using an Evolutionary Algorithm (EA) coupled with an Euler/Navier-Stokes code. To overcome enormous computational time necessary for the design, aerodynamic evaluations are parallelized on Numerical Wind Tunnel (NWT) at National Aerospace Laboratory, a parallel vector machine with 166 processing elements. Parallelization of function evaluations in EA is straightforward and its performance is extremely good since most of computational time is used by flow computations. The design result indicates that the present EA successfully minimizes both the induced drag and the volume wave drag in the given design space.

High Performance Computing in Chemistry

The current use of high performance computers in chemical research encompasses a wide spectrum of subjects and corresponding techniques. Structures and properties of individual molecules are accessible with accurate quantum mechanical methods while macroscopic properties of solids and liquids and interactions between large molecules are derived from sophisticated modelling approaches. Mathematical modelling and optimizations of industrial processes have an increasing impact on productivity. The retrieval and analysis of the vast amounts of genetic information resulting from genome analysis projects pose new computational challenges. Modern computational chemistry clearly has close ties into the neighbouring fields of solid state physics, genetics, computer science, and engineering. Computers have become an established research tool also in the hands of non-experts due to the increasing availability of high performance hardware and sophisticated software packages. The Research Centre Julich is a major contributor to the field of computer-aided chemistry through its own research programs in solid state physics, materials research and soft matter physics. The Central Institute for Applied Mathematics (ZAM) of the Research Centre Julich operates the largest German supercomputer complex with an aggregate peak power of 0.5 TeraFlops. The Cray/SGI hardware is a combination of two massively parallel (T3E-600/512 and T3E900/256) and three parallel vector machines (T90/16 and two J90) with a combined memory of over 100 GByte. Major chemistry software packages installed on these machines are UniChem (DGauss, CADPAC, Gaussian, MNDO), Molpro, GAMESS, CPMD, Amber and GROMOS. HLRZ (Hochstleistungsrechenzentrum, High Performance Computing Centre) is a cooperation of the Research Centre Julich, DESY (Hamburg) and GMD-Forschungszentrum Informationstechnik (Bonn) which makes a major fraction of these computer resources available to academic and industrial users through peer-reviewed projects. Currently about a third of the HLRZ projects and a significant fraction of the Research Centre’s own computational science activities fall into the broader area of chemistry. The rapidly increasing use of the Julich supercomputers by the chemistry community is the background of the initiative of ZAM to organize a scientific conference on the subject of “High Performance Computing in Chemistry”. A particular focus of the meeting was the impact of computational chemistry on industrial research and productivity. The meeting was held 16.-18.2.1998 and brought together about 180 participants from academic and industrial research institutions. With the intention of creating a forum for the exchange of ideas between method developers concentrated in academia and scientists in industrial R&D the program committee had invited 30 renowned speakers from both communities. The invited lecturers impressively demonstrated the multifaceted and interdisciplinary character High Performance Computing in Chemistry

遺伝的アルゴリズムを用いた三次元翼の空力最適化

A Genetic Algorithm (GA) has been applied to optimize a wing aerodynamic shape for generic subsonic transportation aircraft. A three-dimensional compressible Navier-Stokes (N-S) solver is used to evaluate aerodynamic performance of a wing. Although a GA has proved its ability to solve hard optimization problems, a GA coupled with a three-dimensional N-S solver has not been done yet because it needs large computational time even on the latest supercomputer. To overcome this difficulty, Numerical Wind Tunnel at National Aerospace Laboratory, a parallel vector machine with 166 processing elements, was used. In the optimum design obtained from the present GA, the design principles for the wing developed by existing theory and experience are found to be realized. This indicates feasibility of the present approach for the aerodynamic optimization in advanced computational environment.

Challenges in obtaining peak parallel performance with a convex C240, a parallel vector processor

The behavior of the Linpack 300x300 benchmark is examined in the context of a parallel vector machine architecture. Detailed evaluation is performed with respect to the Convex C240. Issues relating to algorithm design and system characteristics are discussed in the context of the Linpack implementation.

Challenges in obtaining peak parallel performance with a Convex C240, parallel vector processor

This report examines the behavior of the Linpack 300×300 benchmark [Dongarra] on a parallel vector machine. It is observed that the performance of several parallel vector machines on this application is far below their nominal peak performance. Dissection of the internals of the algorithms shows how peak performance is limited. The insights gained provide guidance to algorithm developers as to ways to make maximum use of architectural strengths. System architects may gain insight about which system characteristics to optimize to increase the performance of future designs for this class of application.