124 High Performance Visualization
et al. demonstrated good scalability of both indexing and query evaluation,
to several thousands of cores. For details, see the work by Chou et al. [9].
7.3 Formulating Multivariate Queries
Semantic indexing provides the user with the ability to quickly locate
data subsets of interest. In order to make effective use of this ability, efficient
interfaces and visualization methods are needed that allow the user to quickly
identify data portions of interest, specify multivariate data queries to extract
the relevant data, and validate query results. As mentioned earlier, a QDV-
based analysis is typically performed in a process of iterative refinement of
queries and analysis of query results. To effectively support such an iterative
workflow, the query interface and visualization should provide the user with
feedback on possible strategies to refine and improve the query specification,
and be efficient to provide the user with fast, in-time feedback about query-
results, in particular, within the context of large data.
Scientific visualization is very effective for the analysis of physical phenom-
ena and plays an important role in the context of QDV for the validation of
query results. Highlighting query results in scientific visualizations provides an
effective means for the analysis of spatial structures and distributions of the
selected data portions. However, scientific visualization methods are limited
with respect to the visualization of high-dimensional variable space in that
only a limited number of data dimensions can be visualized at once. Scientific
visualizations, hence, play only a limited role as interfaces for formulating
multivariate queries.
On the other hand, information visualization methods—such as scatter-
plot matrix and parallel coordinate plots—are very effective for the visualiza-
tion and exploration of high-dimensional variable spaces and the analysis of
relationships between different data dimensions. In the context of QDV, infor-
mation visualizations, therefore, play a key role as interfaces for the specifica-
tion of complex, multidimensional queries and the validation of query results.
Both scientific and information visualization play an important role in
QDV. In the context of QDV, multiple scientific and information visualization
views—each highlighting different aspects of the data—are, therefore, often
linked to highlight the same data subsets (queries) in a well-defined manner
to facilitate effective coordination between the views. In literature, this design
pattern is often referred to as brushing and linking [34]. Using multiple views
allows the user to analyze different data aspects without being overwhelmed
by the high dimensionality of the data.
To ease validation and refinement of data queries, automated analysis
methods may be used for the post-processing of query results to, for example,
segment and label the distinct spatial components of a query. Information de-
rived through the post-processing of query results provides important means