1 / 32

Introduction to Scientific Computing II

Introduction to Scientific Computing II. Preconditioned CG Continued. Dr. Miriam Mehl. Conjugate Gradients – Basic Idea. solution of SLE minimization iterative one-dimensional minima no repeating search directions. Conjugate Gradients – Principle. CG – Convergence.

bena
Download Presentation

Introduction to Scientific Computing II

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Introduction to Scientific Computing II Preconditioned CG Continued Dr. Miriam Mehl

  2. Conjugate Gradients – Basic Idea • solution of SLE • minimization • iterative • one-dimensional minima • no repeating search directions

  3. Conjugate Gradients – Principle

  4. CG – Convergence • Poisson with 5-point-stencil • like SOR • no parameter adjustment

  5. PCG – Idea convergence rate cg: • Solve system M-1Ax=M-1b • better condition number k • M-1 easy to apply

  6. PCG – Algorithm

  7. PCG – Preconditioner • (Jacobi) • SSOR • incomplete Cholesky decomposition • incomplete LU decomposition • (algebraic) multigrid

  8. Dr. Miriam Mehl Introduction to Scientific Computing II Parallel MG – A Scalable Alternative to Parallel CG?

  9. Can we parallelize a solver iteration in a scalable way? Parallel Solvers – Story 1

  10. parallel cg with “optimal” speedup #processors ~ #unknowns Parallel Speedup CG

  11. parallel smoother with optimal speedup #processors ~ #unknowns (on the finest level!) Parallel Speedup MG

  12. Can we parallelize the solver in a scalable way? Parallel Solvers -- Story 2

  13. parallel cg with “optimal” speedup per iteration #processors ~ #unknowns Parallel Speedup CG

  14. parallel smoother with optimal speedup #processors ~ #unknowns (on the finest level!) Parallel Speedup MG

  15. parallel cg with “optimal” speedup #processors ~ #unknowns Parallel Speedup CG

  16. parallel cg with “optimal” speedup #processors ~ #unknowns Parallel Speedup CG

  17. parallel cg with “optimal” speedup #processors ~ #unknowns Parallel Speedup CG

  18. Parallel Speedup CG h h/2 h/4 h/8 O(1/h) total time to solve? O(ln(1/h)/h)

  19. parallel smoother with optimal speedup #processors ~ #unknowns (on the finest level!) #processors = const on all levels time T time T/4 computation > communication Parallel Speedup MG

  20. parallel smoother with optimal speedup #processors ~ #unknowns (on the finest level!) #processors = const on all levels time T time T time T/2 computation < communication Parallel Speedup MG

  21. parallel smoother with optimal speedup #processors ~ #unknowns (on the finest level!) #processors = const on all levels time T #dof < #proc Parallel Speedup MG

  22. parallel smoother with optimal speedup #processors ~ #unknowns (on the finest level!) #processors = const on all levels time T time T time T #dof < #proc Parallel Speedup MG

  23. Parallel Speedup MG level 0 #dof < #proc level L1 Tcomp < Tcomm level L2 Tcomp > Tcomm level Lmax

  24. Parallel Speedup MG level 0 #dof < #proc level L1 Tcomp < Tcomm level L2 d Tcomp > Tcomm level Lmax

  25. Parallel Speedup MG level 0 #dof < #proc level L1 Tcomp < Tcomm level L2 d Tcomp > Tcomm level Lmax

  26. Parallel Speedup MG level 0 #dof < #proc level L1 Tcomp < Tcomm c level L2 d Tcomp > Tcomm level Lmax

  27. Parallel Speedup MG level 0 #dof < #proc level L1 Tcomp < Tcomm c level L2 d Tcomp > Tcomm level Lmax

  28. Parallel Speedup MG level 0 ln(p) #dof < #proc level L1 Tcomp < Tcomm c level L2 d Tcomp > Tcomm level Lmax

  29. Parallel Speedup MG level 0 ln(p)/2 #dof < #proc level L1 Tcomp < Tcomm c level L2 d Tcomp > Tcomm level Lmax

  30. Parallel Speedup MG p 0 total time per iteration? ln(p) ln(p)+c ln(p)+c+d

  31. Parallel Speedup MG h h/2 h/4 h/8 O(1) O(ln(1/h)) total time to solve?

  32. Iterative Solvers – Overview X system matrix s.p.d. or diagonally dominant and othersX system matrix must be s.p.d.!!!

More Related