An Algebraic Approach to XQuery Optimization

As more data is stored in XML and more applications need to process this data, XML query optimization becomes performance critical. While optimization techniques for relational databases have been developed over the last thirty years, the optimization of XML queries poses new challenges. Query optimizers for XQuery, the standard query language for XML data, need to consider both document order and sequence order. Nevertheless, algebraic optimization proved powerful in query optimizers in relational and object oriented databases. Thus, this dissertation presents an algebraic approach to XQuery optimization. In this thesis, an algebra over sequences is presented that allows for a simple translation of XQuery into this algebra. The formal definitions of the operators in this algebra allow us to reason formally about algebraic optimizations. This thesis leverages the power of this formalism when unnesting nested XQuery expressions. In almost all cases unnesting nested queries in XQuery reduces query execution times from hours to seconds or milliseconds. Moreover, this dissertation presents three basic algebraic patterns of nested queries. For every basic pattern a decision tree is developed to select the most effective unnesting equivalence for a given query. Query unnesting extends the search space that can be considered during cost-based optimization of XQuery. As a result, substantially more efficient query execution plans may be detected. This thesis presents two more important cases where the number of plan alternatives leads to substantially shorter query execution times: join ordering and reordering location steps in path expressions. Our algebraic framework detects cases where document order or sequence order is destroyed. However, state-of-the-art techniques for order optimization in cost-based query optimizers have efficient mechanisms to repair order in these cases. The results obtained for query unnesting and cost-based optimization of XQuery underline the need for an algebraic approach to XQuery optimization for efficient XML query processing. Moreover, they are applicable to optimization in relational databases where order semantics are considered