Abstract
One of the major challenges facing a data warehouse is to improve the query response time while keeping the maintenance cost to a minimum. Recent solutions to tackle this problem suggest to selectively materialize certain views and compute the remaining views on-the-fly, so that the cost is optimized. Unfortunately, in case of a spatial data warehouse, both the view materialization cost and the on-the-fly computation cost are often extremely high. This is due to the fact that spatial data are larger in size and spatial operations are more complex and expensive than the traditional relational operations. In this paper, we propose a new notion, called preview, for which both the materialization and on-the-fly costs are significantly smaller than those of the traditional views. Essentially, to achieve these cost savings, a preview pre-processes the non-spatial part of the query, and maintains pointers to the spatial data. In addition, it exploits the hierarchical relationships among the different views by maintaining a universal composite lattice, and mapping each view onto it. We optimally decompose a spatial query into three components, the preview part, the materialized view part and the on-the-fly computation part, so that the total cost is minimized. We demonstrate the cost savings with realistic query scenarios.