Parallel implementation of a Newton-Krylov flow solver on unstructured grids
A parallel Newton-Krylov flow solver for the Euler equations is presented. The flowfield is solved to steady-state using an inexact-Newton method. An additive-Schwarz preconditioned Generalized Minimal Residual (GMRES) algorithm is used to solve each linear system, with incomplete lower-upper (ILU)...
Uloženo v:
| Hlavní autor: | |
|---|---|
| Médium: | Dissertation |
| Jazyk: | angličtina |
| Vydáno: |
ProQuest Dissertations & Theses
01.01.2004
|
| Témata: | |
| ISBN: | 9780612914186, 0612914186 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | A parallel Newton-Krylov flow solver for the Euler equations is presented. The flowfield is solved to steady-state using an inexact-Newton method. An additive-Schwarz preconditioned Generalized Minimal Residual (GMRES) algorithm is used to solve each linear system, with incomplete lower-upper (ILU) factorization applied locally in each processor. Several grid partitioning algorithms are compared. Ghosting is done for boundary cells and faces. The solver demonstrates good parallelism in both 2D and 3D tests. Parallel scalability is found to be determined by the number of GMRES iterations in 2D and local ILU factorization time in 3D. Parallel performance becomes less sensitive to ILU fill level for large number of processors. A low fill level is recommended for parallel flow solves. The ParMETIS partitioning method performed best in parallel, followed by Reverse Cuthill-McKee reordering and space-filling curve methods. Recommendations are made for further research in this area. |
|---|---|
| Bibliografie: | SourceType-Dissertations & Theses-1 ObjectType-Dissertation/Thesis-1 content type line 12 |
| ISBN: | 9780612914186 0612914186 |