Virtual Partitioning ad-hoc Queries over Distributed XML Databases

Abstract

XML query processing on large repositories suffers from performance issues. Despite many efficient indexing techniques, oftentimes only physical XML data fragmentation techniques can improve query processing performance. In such approaches, the database is physically partitioned based on the attributes and selection criteria used in the most frequent queries of the system. Distributed query processing can then take advantage of pruning irrelevant fragments and processing the relevant ones in parallel. However, in many applications, such as Decision Support Systems, input queries are ad-hoc. In such cases, there is no frequent attribute access and physical partitioning is not an option. In relational settings, virtual partitioning has been successfully used to improve performance in such scenario with parallel query processing. Inspired by those solutions, in this work we adapt the relational virtual partitioning technique to the XML model, and apply it to improve the performance of ad-hoc analytical XML queries. Our experimental analysis shows the effectiveness of our approach.