Access methods for xml query optimization
Attila Barta · 2006
An ever increasing number of organizations store their data using the XML data model. In this respect, the XML data model is used as encoding for information exchange, e-commerce messaging and complex data formats such as genome and protein data. These XML encoded data are stored in XML documents and a collection of XML documents is know as an XML database. In a simplified model, an XML document can be modeled as a tree. A challenge to the current database research is how to query these tree-encoded databases. The de facto XML query language is XQuery and there are many XQuery implementations ranging from simple XQuery prototypes to complex implementations that are part of database systems that provide XML support. The common denominator of these XQuery implementations is the immaturity of the XML query optimizer. The query optimization techniques proved to be the foundation of success for the object/relational database systems. We believe that without a sound query optimization technique querying large XML databases will not be feasible. In this thesis we present query optimization techniques for XML databases. These query optimization techniques are based on an original query optimization paradigm that we call two-level query optimization and in the context of this paradigm we introduce two novel optimization strategies, namely holistic path summary pruning and access-order selection. Furthermore, we propose the usage of path summaries (a concise representation of an XML instance) in the query optimizer as system catalogs. The query optimization techniques that we propose are employed at the access method selection level. An access method is a query operator that accesses a data sources. In this thesis we present three access methods: ToXStream, which works on streamed documents, TwigStackScan and ToXinScan that employ path summaries. The soundness of the optimization techniques that we propose are verified through experiments on XML query benchmarks. The experimental results suggest that, when applicable, these optimization techniques induce performance improvements up to two orders of magnitude over the state-of-the-art.