Adaptive XML-to-Relational Storage Strategies

Background

The main concern of the database-based XML techniques is the choice of the way XML data are stored into relations, so-called XML-to-relational mapping or schema decomposition to relations. The strategy in first approaches to XML-to-relational mapping, so-called generic (e.g. Florescu et al., 1999), was based purely on the data model of XML documents. The methods were able to store any kind of XML data since they viewed XML documents as general labelled trees. But, the efficiency of query evaluation was quite low due to numerous join operations or the increase of efficiency was gained at the cost of increase of space overhead.

Hereafter, the scientists came with a natural idea to exploit structural information extracted from XML schemas of XML data, usually expressed in DTD (Document Type Description) (Bray et al., 2006) or XML Schema (Thompson et al., 2004; Biron et al., 2004) language. All the so-called schema-driven approaches (e.g. Shanmugasundaram et al., 1999) were based on the same idea that the structure of the target relational schema can be created according the structure of the source XML schema. Assuming that a user specifies the XML schema as precisely as possible to specify the related data, we can get also more precise relational XML schema. The problem is that DTDs are usually too general. The extensive examples are recursion or * operator which, in general, enable to specify infinitely deep or wide XML documents. According to analyses of real-world XML data (Mlynkova et al., 2006) in both the cases the respective XML documents are much simpler and, thus, the effort spent on processing all the complex schema constructs is useless.