Database Reference
In-Depth Information
to scientific data management: they must support a plethora of input data
formats, and therefore, they contain a number of data loader modules; they
must create an internal data structure that is suitable for use by a potentially
large collection of visualization, analysis, and rendering modules, all of which
may potentially run in parallel on shared or distributed memory machines. An
open problem is one of data models and semantics, where meaning is assigned
to arrays of data stored in data files.
With data of massive scale, it is often useful to perform a multiresolution
analysis, working first with a smaller, coarser version of data, then progres-
sively refine the analysis as interesting features are revealed. We saw that a
space-filling curve model has proven to be highly ecient for interactive anal-
ysis of massive data. However, such a data model and layout is unlikely to be
output directly from a simulation. This is a good example of how a data model
and layout that works very well for multiresolution analysis is unlikely to be
used by simulations for output. Multiresolution, quantitative feature detection
and analysis methods were demonstrated on two different datasets from the
field of turbulent mixing. This quantitative analysis approach is very useful for
enabling scientific knowledge discovery by focusing on features rather than on
the machinery for creating potentially incomprehensible images of large-scale
scientific data.
A significant barrier faced by many computational and experimental sci-
ence projects is the complexity of using state-of-the art technology from sci-
entific data management. We discussed an approach for encapsulating com-
plexity in the form of a high-level API for data storage and retrieval that
lowers the entry point for using such technology. This concept was also ap-
plied to index/query technology. We have applied these concepts to multi-
ple application areas to produce results showing that visual data analysis,
as a field, can benefit from a close collaboration with the field of scientific
data management. The benefit is improved performance for many data in-
tensive operations, like data I/O and data subsetting, as well as the poten-
tial to conceive and create completely new, paradigm-changing approaches
to solve the problem of scientific knowledge discovery for massive, complex
datasets.
Acknowledgments
This work was supported by the Director, Oce of Advanced Scientific Com-
puting Research, Oce of Science, of the U.S. Department of Energy, under
Contract No. DE-AC02-05CH11231 through the Scientific Discovery through
Advanced Computing (SciDAC) program's Visualization and Analytics Cen-
ter for Enabling Technologies (VACET). This research used resources of the
