Estimating the Selectivity of XML Path Expressions for Internet Scale Applications

Data on the Internet is increasingly presented in XML format. This enables novel applications that pose queries over “all the XML data on the Internet.” Queries over XML data use path expressions to navigate through the structure of the data, and optimizing these queries requires estimating the selectivity of these path expressions. In this paper, we propose two techniques for estimating the selectivity of simple XML path expressions over complex large-scale XML data as would be handled by Internet-scale applications: path trees and Markov tables. Both techniques work by summarizing the structure of the XML data in a small amount of memory and using this summary for selectivity estimation. We experimentally demonstrate the accuracy of our proposed techniques, and explore the different situations that would favor one technique over the other. We also demonstrate that our proposed techniques are more accurate than the best previously known alternative.