Section: New Results

Advanced Algorithms for Efficient XML processing

In 2013, several research works of the team focusing on advanced algorithms for processing XML data have been finalized and concluded through prestigious journal publications.

A first line of work concerned the usage of materialized views to speed up the evaluation of complex XML queries. In our previous work we had demonstrated that such views may bring up very significant speed-up factors of several orders of magnitude. However, materialized views need to be kept up to date when the underlying database changes. In [14] we have described efficient algorithms for updating materialized views expressed in a rich dialect of XQuery, the standard query language for XML.

A second class of work was concerned with XML static type analysis, in particular with the crucial problem of deciding XML type inclusion, that is: whether any XML tree of type ${\tau }_1$ is also of type ${\tau }_2$ where ${\tau }_1,{\tau }_2$ are XML types with interleaving and counting (currently adopted by main stream schema languages). For these types, inclusion is EXPSPACE-complete. We have defined and formally studied a quadratic subtype-checking algorithm for the case where the right-hand side type ${\tau }_2$ meets some restrictions on symbol occurrences and the use of counting. These restrictions are often met by human-designed types, so our technique perfectly fits the needs of typical XML type-checking algorithms, which frequently require to check for inclusion a machine-generated subtype ${\tau }_1$ against a human-defined supertype ${\tau }_2$. Our approach has been validated by extensive experimental results [16] . In addition, we have devised and formally studied an alternative algorithm, still for the asymmetric case where ${\tau }_2$ is restricted, based on structural, top-down analysis of types expression. This algorithm is almost linear: it has a linear-time backbone, and resorts to the above quadratic approach for some specific parts of the compared types. Our experiments show that this new algorithm is much faster than the quadratic one and that it typically runs in linear time, hence it can be used as a building block for a practical type-checking compiler for XML programs and queries [15] .

Third, we have completed and concluded our work on type-based document projection for efficient XML data management. The idea here is to restrict XML documents, prior to evaluating a query over them, to only those parts of the document that the query actually needs to consult. We provide algorithms for determining such document parts and experimentally demonstrate the benefits of such techniques, in [13] .

Finally, we have devised a system that is able to process both queries and updates on very large XML documents [22] . As observed in recent works, such very large documents are generated and processed in several contexts, in particular in those involving scientific data and logs. Our system supports a large fragment of XQuery and XUF (XQuery Update Facility). The system exploits dynamic and static partitioning to distribute the processing load among the machines of a MapReduce cluster. The proposed technique applies when queries and updates are iterative, i.e., they iterate the same query/update operations on a sequence of subtrees of the input document. From our experience many real world queries and updates actually meet this property. Our partitioning technique is schema-less, as the presence of a user-supplied schema is not required; indeed, this technique only relies on path information extracted from the input query/update. Experiments conducted on a 8-machine Hadoop cluster have demonstrated that the system is able run both iterative queries and updates on quite large documents.