Cost Estimation Techniques for Database Systems

This dissertation is about developing advanced selectivity and cost estimation techniques for query optimization in database systems. It addresses the following three issues related to current trends in database research: estimating the cost of spatial selections, building histograms without looking at data, and estimating the selectivity of XML path expressions. The first part of this dissertation deals with estimating the cost of spatial selections, or window queries, where the query windows and the data objects are general polygons. Previously proposed cost estimation techniques only handle rectangular query windows over rectangular data objects, thus ignoring the significant cost of exact geometry comparison (the refinement step in a “filter and refine” query processing strategy). The cost of the exact geometry comparison depends on the selectivity of the filtering step and the average number of vertices in the candidate objects identified by this step. We develop a cost model for spatial selections that takes these parameters into account. We also introduce a new type of histogram for spatial data that captures the size, location, and number of vertices of the spatial objects. Capturing these attributes makes this type of histogram useful for accurate cost estimation using our cost model, as we experimentally demonstrate. The second part of the dissertation introduces self-tuning histograms. While similar in structure to traditional histograms, self-tuning histograms are built not by examining the data or a sample thereof, but by using feedback from the query execution engine about the selectivities of range selections on the histogram attributes to progressively refine the histogram. Since self-tuning histograms have a low up-front cost and the cost of building them is independent of the data size, they are an attractive alternative to traditional histograms, especially multidimensional histograms. The low cost of self-tuning histograms can help a self-tuning selfadministering database system experiment with building many different histograms on many different combinations of data columns. This is useful since the system cannot rely on a database administrator to decide which histograms to build.