AVBp - AVBP Parallel Computing
- Author
: Thilo Schönfeld, CERFACS
- schonfel@cerfacs.fr
Objectives :
This project centers around parallel computing aspects in CFD.
The computations are based on the unstructured flower solver
AVBP
of CERFACS. Two tasks are proposed:
- Parallel computing:
execution of AVBP on 1, 2, 4, 8, etc. processors, measuring of CPU
times and creation of a "speed-up" table. The influence on
the restitution time by parameters such as the mesh size,
the numerical methods or the partitioning algorithms is investigated.
- Programming exercise:
routines are written in FORTRAN 77/90 language that
allow to calculate (in a parallel and unstructured code environment)
the minimal wall distance for each grid point.
The computation of these minimal wall distances is useful when for
example using wall functions for turbulence models.
A strategy is to be established to obtain this information on the
nearest wall distance across several processors.
Working plan :
- AVBp.1 : Introduction to the AVBP flow solver, execution of an academic
test case (channel with bump).
- AVBp.2 : Parallel computing (speed-up, etc.) on a multiple processor
machine, grid size influence, etc.
- AVBp.3 : Conceiving a basic programming
strategy for the nearest wall-distance calculation,
programming of FORTRAN routines and application to test cases.
|
Week |
1 |
2 |
3 |
4 |
5 |
6 |
|
AVBp.1 |
**** |
* |
|
|
|
|
|
AVBp.2 |
|
*** |
* |
|
|
|
|
AVBp.3 |
|
|
*** |
**** |
**** |
|
|
Exam |
|
|
|
|
|
**** |
Deliverables :
- 1. Speed-up tables
- 2. FORTRAN routines for wall-function calculation
- 3. Report
Bibliography :
- [1] The AVBP User's Manual, T. Schönfeld, CERFACS, 1999.
- [2] Cours de FORTRAN77/90.
Pedagogical interest :
Introduction to parallel computing.
Back to
index