Database Reference
In-Depth Information
Visualization
Slicing, contouring, volume rendering...
Raw Data
Re-ordered Data
E.g.
Hierarchical Z-order
Brick decomposition
Statistics
Analysis
Topological feature analysis,
feature quantification, tracking...
Figure 9.3
Hierarchical Z-ordered data layout and topological analysis com-
ponents highlighted in the context of a typical visualization and analysis
pipeline.
provided by the Z-order space filling curve by incorporating a coarse-to-fine
hierarchy on the ordering. Our system is very simple to implement and has
been used as a core technology and applied to a variety of visualization algo-
rithms, such as slicing, isosurfacing, and volume rendering on massive amounts
of scientific simulation data.
A multiresolution data layout is a key technology that plays a central role in
advanced analysis algorithms that go beyond simple images or movies. Topo-
logical analysis is one such technique that is useful for providing a deeper un-
derstanding of scientific phenomena. In this type of application, the analysis
takes the form of defining, detecting, and quantifying features in data. Fea-
tures can and do exist at multiple scales in data. For this reason, an ecient,
multiresolution data layout and model is an integral part of high-performance
implementations of such algorithms. For more information on state-of-the art
topological analysis, see,
17
-
22
and for application of such techniques to the
analysis of simulation data, including hydrodynamic instability, and compar-
ative analysis.
23
-
25
9.3.2 Hierarchical Indexing for Out-of-Core Access
to Multiresolution Data
Out-of-core computing
26
specifically addresses the issues of algorithm redesign
and data layout restructuring that are necessary to enable data access patterns
having minimal out-of-core processing performance degradation. Research in
this area is also valuable in parallel and distributed computing, where one
has to deal with the similar issue of balancing processing time with the time
required for data access and movement among elements of a distributed or
parallel application.
The solution to the out-of-core processing problem is typically divided into
two parts: (1) algorithm analysis, to understand data access patterns and,
when possible, redesign to maximize data locality; (2) storage of data in sec-
ondary memory using a layout consistent with the access patterns of the
