Estimating Answer Sizes for XML Queries

Abstract

Estimating the sizes of query results, and intermediate results, is crucial to many aspects of query processing. In particular, it is necessary for effective query optimization. Even at the user level, predictions of the total result size can be valuable in “next-step” decisions, such as query refinement. This paper proposes a technique to obtain query result size estimates effectively in an XML database.

Queries in XML frequently specify structural patterns, requiring specific relationships between selected elements. Whereas traditional techniques can estimate the number of nodes (XML elements) that will satisfy a node-specific predicate in the query pattern, such estimates cannot easily be combined to provide estimates for the entire query pattern, since element occurrences are expected to have high correlation.

We propose a solution based on a novel histogram encoding of element occurrence position. With such position histograms, we are able to obtain estimates of sizes for complex pattern queries, as well as for simpler intermediate patterns that may be evaluated in alternative query plans, by means of a position histogram join (pH-join) algorithm that we introduce. We extend our technique to exploit schema information regarding allowable structure (the no-overlap property) through the use of a coverage histogram.

We present an extensive experimental evaluation using several XML data sets, both real and synthetic, with a variety of queries. Our results demonstrate that accurate and robust estimates can be achieved, with limited space, and at a miniscule computational cost. These techniques have been implemented in the context of the TIMBER native XML database [22] at the University of Michigan.

H.V. Jagadish and Yuqing Wu were supported in part by NSF under grant IIS- 9986030 and DMI-0075447. Jignesh M. Patel was supported in part by a research gift donation from NCR Corporation.