Cost modeling of spatial operators using non-parametric regression

In an object-relational database management system, a query optimizer requires users to provide cost models of user-defined functions. The traditional approach is analytical, that is, it builds a cost model generated as a result of analyzing the query processing steps. This analytical approach is difficult, however, especially for spatial query operators because of the complexity of the processing steps. In this paper, a new approach that uses non-parametric regression is proposed. This approach significantly simplifies the process of building a cost model, while achieving highly accurate cost estimation. We demonstrate the simplicity and efficacy of this approach through experiments for three spatial operators—the range query, the window query, and the k-nearest neighbor query—commonly used in spatial databases, using both real and synthetic data sets.