An Efficient Index Lattice for XML Query Optimization

Structural indexes of XML data can effectively reduce the search space for the evaluation of path queries over the data. The indexes partition the structural graph of an XML document into equivalent classes of nodes that are then condensed into index nodes. However, structural indexes are inadequate to handle queries with valuebased conditions, since equivalent nodes in the same partition become distinguishable by their data contents. In practice, only a small portion of nodes in each partition are relevant for the processing of a value-based condition. To enhance the applicability of structural indexes, we propose a lattice structure on an XML structural index, which we call the Structure Index Tree (SIT). The index is defined as partitions of equivalent paths in an XML document, while an element in the lattice, which we call a SIT-Lattice Element (SLE), is an index of an arbitrary subset of paths in the document. Since paths represent the structure of XML data and each text node is associated with a unique path, we can define an SLE to filter out both irrelevant structures and text nodes. We propose a set of SLE operations and devise efficient techniques to generate SLEs that can be tailored towards query workloads. Our experiments show that SLEs significantly speed up the evaluation of path queries with value-based and aggregation-based conditions. We also demonstrate that SLEs are able to support effective querying over very large XML documents in memory-limited hand-held devices.