Domain Decomposition Parallelization of Mesh Based Applications

11.01.2002


Start Lecture by clicking here


Content

Author:Panagiotis Adamidis and Thomas Bönisch

  1. Domain Decomposition Parallelization of Mesh Based Applications
  2.     Outline
  3.     Parallelization - Target
  4.     A Problem (I)
  5.     Parallelization strategies
  6.     Parallelization Problems
  7.     Concepts - Message Passing (II)
  8.     How to split the Domain in the Dimensions (I)
  9.     How to split the Domain in the Dimensions (II)
  10.     Replication versus Communication (I)
  11.     Replication versus Communication (II)
  12.     2- Dimensional DD with two Halo Cells
  13. Example: Parallelization of a 3D Finite Volume Flow Solver
  14.     Starting Point: Sequentiell Program
  15.     Dynamic Data Structures
  16.     Main Loop
  17.     Heptadiagonalmatrix
  18.     Parallelization - Solver (I)
  19. Parallelization - Solver (II)
  20.     Parallelization - Solver (III)
  21.     A Real Heptadiagonalmatrix
  22.     Difference between strong / weak coupling
  23.     Comparison of Solvers - Convergence
  24.     Comparison of Solvers - Performance
  25.     Results - Solving Method
  26.     Domain Decomposition
  27.     Speedup on Cray T3E
  28.     Scaleup on Cray T3E
  29.     Domain Decomposition of Unstructured Grids
  30.     Unstructured FEM Grid with Global Numbering
  31.     Shape of Corresponding System Matrix
  32.     Domain Decomposition
  33.     1. Nonoverlapping Domain Decomposition
  34.     2. Grid Points Separated into Inner and Boundary Points
  35.     3. Renumbering of the Inner and Boundary Points
  36.     Arrow Shaped System Matrix after Renumbering
  37.     Data Distribution
  38.     Linear System
  39.     Direct Substructuring
  40.     Transformations of the Original System
  41.     Schurcomplement System
  42.     1.Direct Factorization
  43.     2.Assembiing of the Schur complement system (I)
  44.     2.Assembiing of the Schur complement system (II)
  45.     3.Solving of the Schur complement system
  46.     4.Solving of the interior unknows
  47.     Parallel Computations on the Subdomains
  48.     Overlapping Domain Decomposition
  49.     Literature

Back to the Parallel Programming Workshop Overview