Efficient Database Support for Spatial Applications
Monday, April 19, 1999
11:00-12:15
Interschool Lab, CEPSR
Abstract
Database technology had come a long way since its inception nearly four
decades ago when the relational data model was invented to ease the task of
managing data. Relational database systems have proven their effectiveness
for managing large volumes of alphanumeric data, and are now a critical part
of most organizations' operating cycle. However, relational database
systems are not very effective for storing and querying data of complex
types like images, text and spatial data. For the past ten years a great
deal of research has focused on building database systems based on an
"object-relational" model that allows processing data of complex types. A
large fraction of this research has been directed towards building
object-relational database systems that can handle geo-spatial workloads.
While researchers have always acknowledged the existence of very large data
sets in the geo-spatial domain, the vast majority of research to date has
focused on language issues or uniprocessor query evaluation and indexing
techniques. This is unfortunate, since the large data set problems that
have been lurking beneath the surface are now likely to surface with a
vengeance.
This talk describes techniques for building a parallel scalable geo-spatial
database system. The first part of this talk focuses on efficient algorithms
for evaluating spatial operations in a centralized database system. The
second part of the talk focuses on using parallelism to evaluate spatial
operations. This part addresses two issues: how to decluster data in a
parallel environment, and how to use parallelism to evaluate spatial
operations. Performance results from an actual implementation will also be
presented to support the feasibility of these techniques.
Chris Okasaki
cdo@cs.columbia.edu