Fractional XSketch Synopses for XML Databases

Abstract

A key step in the optimization of declarative queries over XML data is estimating the selectivity of path expressions, i.e., the number of elements reached by a specific navigation pattern through the XML data graph. Recent studies have introduced XSketch structural graph synopses as an effective, space-efficient tool for the compile-time estimation of complex path-expression selectivities over graph-structured, schema-less XML data. Briefly, XSketches exploit localized graph stability and well-founded statistical assumptions to accurately approximate the path and branching distribution in the underlying XML data graph. Empirical results have demonstrated the effectiveness of XSketch summaries over real-life and synthetic data sets, and for a variety of path-expression workloads.

In this paper, we introduce fractional XSketches (fXSketches) a simple, yet intuitive and very effective generalization of the basic XSketch summarization mechanism. In a nutshell, our fXSketch synopsis extends the conventional notion of binary stability (employed in XSketches) with that of fractional stability, essentially recording more detailed path/branching distribution information on individual synopsis edges. As we demonstrate, this natural extension results in several key benefits over conventional XSketches, including (a) a simplified estimation framework, (b) reduced run-time complexity for the synopsis-construction algorithm, and (c) lifting the need for critical uniformity assumptions during estimation (thus resulting in more accurate estimates). Results from an extensive experimental study show that our fXSketch synopses yield significantly better selectivity estimates than conventional XSketches, especially in the context of complex path expressions with branching predicates.