Title: XML engines: main-memory versus relational-like systems.
Nicole Bidoit-Tollu, Dario Colazzo · 2009
• relational-like engines assume that the full document is represented by means of several tables, each of which contains information (identifier plus parent-child foreign keys) about all elements of a given tag. Tables are stored on the disk, and loaded for query evaluation according to some particular strategy. Main-memory XML engines are obviously limited by main-memory consumption and can not process large documents. This is the starting point of optimization techniques for query and update evaluation based on document projection. These techniques [1,2,3] are based on pruning the document during the loading phase, by keeping only the parts needed for the evaluation. Pruning is done before query/update evaluation, in order to reduce the main-memory space necessary for the evaluation. Concerning relational-like engines, in some cases they avoid loading useless tables during query evaluation, by following strategies commonly adopted in relational DBMSs, thus having the possibility of processing quite large documents. The main goal of the project is to carry out a comparative analysis between main-memory and table-baed engines, under the constraint that main-memory engines are optimized by projection based techniques. In particular, the objective is to characterize classes of queries and type of documents for which main-memory systems with projection have better/worse performances than relational-like systems.