Currently, we are focussing on improvements to the second of the three components. Most sophisticated linear solvers, for example, multigrid solvers or parallel solvers, are rather complicated and difficult to use. Our approach is to build a knowledge-based system that allows the information stored in the element stiffness matrices, generated by the first component, to be provided to various solver libraries in the correct format for use by those libraries. Data partitioning for parallel execution is included within this process. The approach is not limited to linear analyses but can also be applied for nonlinear problems.

In our presentation, we will talk about the program structure of our tool and in particular on the flexible approach by which additional solver libraries may be included.

Guy Lonsdale is a senior research staff member at the C & C Research Laboratory. His area of activities is parallel algorithms and applications. Current activities include investigations of the use of the HPF approach for the crashworthiness code PAM-CRASH, contributing to the ESPRIT Long Term Research project HPF+ as part of a collaboration with ESI.