Unstructured Data Partitioning for Large Scale Visualization

1
Unstructured Data Partitioning for Large Scale
Visualization

• CSCAPES Workshop
• June, 2008
• Kenneth Moreland
• Sandia National Laboratories

Sandia is a multiprogram laboratory operated by
Sandia Corporation, a Lockheed Martin
Company,for the United States Department of
Energys National Nuclear Security
Administration under contract DE-AC04-94AL85000.
2
The Parallel Visualization Pipeline
3
The Parallel Visualization Pipeline
4
Data Parallel Pipelines

• Duplicate pipelines run independently on
  different partitions of data.

5
Data Parallel Pipelines

• Duplicate pipelines run independently on
  different partitions of data.

6
Data Parallel Pipelines

• Some operations will work regardless.
• Example Clipping.

7
Data Parallel Pipelines

• Some operations will work regardless.
• Example Clipping.

8
Data Parallel Pipelines

• Some operations will work regardless.
• Example Clipping.

9
Data Parallel Pipelines

• Some operations will have problems.
• Example External Faces

10
Data Parallel Pipelines

• Some operations will have problems.
• Example External Faces

11
Data Parallel Pipelines

• Ghost cells can solve most of these problems.

12
Data Parallel Pipelines

• Ghost cells can solve most of these problems.

13
Data Partitioning

• Partitions should be load balanced and spatially
  coherent.

14
Data Partitioning

• Partitions should be load balanced and spatially
  coherent.

15
Data Partitioning

• Partitions should be load balanced and spatially
  coherent.

16
The Parallel Visualization Pipeline
17
Parallel Rendering
18
Parallel Rendering
19
Tiled Displays
20
Rendering Translucent Geometry
21
Unstructured Volume Rendering in Parallel
22
Unstructured Volume Rendering in Parallel
23
Unstructured Volume Rendering in Parallel
24
Unstructured Volume Rendering in Parallel
25
Unstructured Volume Rendering in Parallel
26
Mesh Partitioning
27
Partitioning on Spatial Structure K-D Tree
28
K-D Trees Provide Query Structures
What elements are closest to here?
29
K-D Trees Provide Query Structures
What regions / elements intersect this view
frustum?
30
K-D Trees Provide Query Structures
8
4
What is the visibility order of the regions from
this viewpoint?
7
1
5
6
2
3
31
Reconstructing Connectivity Information
May not be unique.
Neighbor info usually missing.
32
Reconstructing Connectivity Information
33
Future Work

• Code Optimization and Cleanup
• Integration of other partitioning algorithms.
• Better Data Type Support.
• Better Temporal Support.