19 June 2001



K. Shaw
Naval Research Laboratory
Stennis Space Center, MS 39529-5004
C. Benetz
Tulane University
New Orleans, LA


In spatial databases, information is naturally distributed due to the nature of the collection process of the data as well as the size of the data. In this ad hoc distribution, the spatial objects are not allocated to nodes in order to improve performance of the database. Few papers have addressed the issue of distribution of spatial data for performance reasons [Abel], though research has been done on how to manipulate spatial data that is already distributed.

Distribution of spatial data presents challenges that do not exist in distributing ordinary data. Spatial objects are more complex than ordinary objects. They can be composed of a single object or a series of objects (for example, an area can be defined by a series of connected lines). This makes the use of an object oriented database, to hold and manage the spatial information appealing. In distributing objects over several machines, the areas of horizontal fragmentation, vertical fragmentation, and replication need to be addressed. This is more complicated than the fragmentation and replication of relational data or ordinary objects because of the complexity of the objects.

Not only is spatial data more complex, but the queries that are performed on spatial data are a superset of regular SQL. The queries include a number of geometric operations, such as 'contains' and 'nearest'. These additional operations make more options apparent for distribution, therefore making the decision about how to distribute the spatial objects more difficult.

In this paper, the issues involved in distribution of spatial objects will be discussed, including how objects in general should be distributed. Additionally, preliminary results from experimentation in distribution of spatial objects will be presented. These will be compared to results of experiments in distributing ordinary objects (those that do not contain spatial attributes).

Sponsored by the Marine Corps Warfighting Lab, the Defense Modeling and Simulation Office, and the National Imagery and Mapping Agency s Terrain Modeling Program Office.

Presented at the First Southern Symposium on Computing (SSC 98), Hattiesburg, MS in December 4-5, 1998.
Naval Research Laboratory Contribution Number NRL/PP/7441 98-0016.
Conference Proceedings