The Design and Evaluation of a High-Performance Earth Science
Database
Carter Shock,
Chialin Chang,
Bongki Moon,
Anurag Acharya,
Larry Davis,
Joel Saltz.
Alan Sussman,
To appear in the special issue of Parallel Computing on Parallel
Data Servers and Applications
Abstract:
Earth scientists have encountered two major obstacles in their attempts to
use remotely sensed imagery to analyze the earth's land cover
dynamics. First, the volume of data involved is very large and second,
significant preprocessing is needed before the data can be used. This is
particularly so for studies that analyze global trends using data sets that
cover multiple years. In this paper, we present the design of an earth
science database as well as our early experiences with it. The primary
design goal of this database is to facilitate efficient access to and
preprocessing of large volumes of satellite data.
Our initial design assumed that the main bottleneck in the system would be
retrieving data from the disks. However, experimental results show that
precise identification of all the data values corresponding to a query can
take a significant amount of time. The problem is even more pronounced in
designing the system to attempt to minimize time spent performing I/O. We
therefore discuss a major redesign of the system that includes a reworking
of the indexing scheme and a reorganization of the data on disks.
Preliminary experimental results show that the redesigned system performs
significantly better than the original system, providing interactive
response times for local queries.
Postscript
(compressed 178K)