Author(s)

S. Wuppalapati, A. D. Belegundu, A. Aziz & V. Agarwala

Abstract

Generating Pareto surfaces is a well-accepted technique in multi-attribute decision making. For computationally intensive applications like finite element based optimization, it can become very expensive to generate the complete Pareto surface. Hence, using parallel computer clusters in these scenarios becomes very attractive. Pareto surfaces are generated using two different clusters with a structural topology problem as a test problem and the performance gains realized are analyzed. Near linear speed-ups and high efficiencies are observed on both the clusters. It is possible to integrate this methodology into commercial software applications, leading to less turn around times to make critical decisions in various applications. Keywords: topology optimization, multi-attribute optimization, finite elements, Pareto surfaces, parallel clusters, MPI, speed-up. 1 Introduction Multi-attribute or multi-criteria optimization typically involves the problem of ‘simultaneously’ minimizing (or maximizing) different objectives within a given domain. Mathematically it involves solving the optimization problem: 1 2 3 min [ ( ), ( ), ( ), , ( )] subject to m f f f f ∈Ω x x x x x … (1)

Keywords

topology optimization, multi-attribute optimization, finite elements,Pareto surfaces, parallel clusters, MPI, speed-up.