Grid middleware for engineering design search and optimisation.

Description/Abstract

Design search and optimisation algorithms can be used by engineers to yield improved designs. Design search involving the analysis of the aerodynamic properties of a design, using Computational Fluid Dynamics, is both computationally and data intensive, making
this problem well matched to Grid computing. Evaluation of the quality of a design (the objective function) may require commercial and user supplied software packages to be called in sequence, data transferred to and from suitable compute resources, in addition to preand
post-processing. To allow engineers to express these necessarily complex workflows we expose a suite of Grid-enabled tools to a high-level scripting language. These tools include client functionality to Globus compute resources, and to a job submission Web service which exposes a cycle-scavenging Condor pool. These tools are exposed as functions to the Matlab environment that can be used directly by the engineer, or intergrated into higher level functions for design search and optimisation. Here the benefits of the scripting approach are discussed, together with details of the Grid middleware used and the implementation of the client functionality.