Spatial Indexing with a Scale Dimension

Abstract

It is frequently the case that spatial queries require a result set of objects whose scale — however this may be more precisely defined — is the same as that of the query window. In this paper we present an approach which considerably improves query performance in such cases. By adding a scale dimension to the schema we make the index structure explicitly “aware” of the scale of a spatial object. The additional dimension causes the index structure to cluster objects not only by geographic location but also by scale. By matching scales of the query window and the objects, the query then automatically considers only “relevant” objects. Thus, for example, a query window encompassing an entire world map of political boundaries might return only national borders. Note that “scale” is not necessarily synonymous with “size”. This approach improves performance by both narrowing the initial selection criteria and by eliminating the need for subsequent filtering of the query result. In our performance measurements on databases with up to 40 million spatial objects, the introduction of a scale dimension decreased I/O by up to 4 orders of magnitude. The performance gain largely depends on the object scale distribution.

We investigate a broad set of parameters that affect performance and show that many typical applications could benefit considerably from this technique. Its scalability is demonstrated by showing that the benefit increases with the size of the query and/or of the database. The technique is simple to apply and can be used with any multidimensional index structure that can index spatial extents and can be efficiently generalized to three or more dimensions. In our tests we have used the BANG index structure.