Relational Approach to XPath Query Optimization

This thesis contributes to the Pathfinder project which aims at creating an XQuery compiler on top of a relational database system. Currently, it is being implemented on top of MonetDB, a main memory database system. For optimization and portability purposes, Pathfinder first compiles an XQuery expression into its own relational algebra, before translating the query into the query language of the underlying system (in this case MIL, the MonetDB Interpreter Language). This thesis focusses on the optimizability aspect of the algebra. Pathfinder’s relational algebra is built on a set of basic relational (comparable to SQL) operators extended with the staircase join operator, an operator specifically designed for evaluation of XPath axis steps. By formally specifying the relational model of Pathfinder and all operators of its algebra, this thesis shows how they can be exploited for optimization. The main focus is on the staircase join operator, which, as it seems, does not provide enough rewriting flexibility in its current implementation. Therefore, a new staircase join definition is proposed: the symmetric staircase join.