|
SSP Project Summary:
|
 |
Student
Klaus Aehlig, Ludwig Maximilian University Munich
Supervisor
Michael Bowers, EPCC
The parallelisation of a Finite Element (FE) code called Thermique is
currently in process for the Rho consortium under the UK's High
Performance Computing Initiative. The parallel code is targeted
for the T3E and the original direct linear algrebra solver has been
replaced with an iterative solver with a simple preconditioner.
Due to the T3E's distributed memory, domain decomposition is required
and this can require significant amounts of additional time and effort
to do.
The aim of the project is to parallelise the Thermique code on a
shared memory platform using the OpenMP standard, using a suitable
iterative method.
This project may provide an opportunity to collaborate with project
SS-99-06, which also aims to use OpenMP to parallelise another
iterative method but in a code with a different coefficent matrix
storage method. There could be a beginning of an iterative library
using OpenMP.
There are several reasons for proposing this project:
- Gain general experience of OpenMP
- FE codes have the same typical structure so experience gained can be
used in other FE projects.
- I suspect industrial firms using FE will probably purchase shared
memory machines in future rather than distributed memory ones.
- Potential usefulness to Rho.
The final report for this project is available here.
Webpage maintained by
mario@epcc.ed.ac.uk