Adaptive XML Shredding: Architecture, Implementation, and Challenges

As XML data becomes central to business-critical applications, there is a growing need for efficient and reliable XML storage. Two main approaches have been proposed for storing XML data: native and colonial systems. Native systems (e.g., [9, 20]) are designed from the ground up specifically for XML and XML query languages. Colonial systems (e.g., [5, 7, 19]), on the other hand, attempt to reuse existing commercial database systems (DBMS) by mapping XML into the underlying model used by the DBMS. Colonial systems can thus leverage features, such as concurrency control, crash recovery, scalability, and highly optimized query processors available in the DMBS, making them an attractive alternative for managing XML data. However, several technical challenges need to be addressed in terms of architecture, algorithms, and implementation of these systems. In this paper, we described how these issues are addressed in the context of colonial systems that use relational databases as the underlying DBMS. The mismatch between the XML and the relational models implies that one must first shred an XML tree-structured document so that it fits into flat relational tables. Therefore, a mechanism is needed to determine the appropriate storage configuration. Once a mapping is selected, the system must provide support for loading the XML data into the database, and to translate queries over the original document into queries over the mapped data. There are different approaches for these problems. For example, while commercial relational systems require users to manually define mappings [14, 15], techniques have been proposed to automatically derive XML-to-relational mappings that adopt either a fixed shredding strategy [19, 11] or that derive the best shredding for a given application [5, 4]. Different techniques have also been proposed for query translation [10, 6]. Although individual problems pertaining to colonial XML storage systems have been studied in isolation, to the best of our knowledge, the design and implementation of a complete colonial system has not been described in the literature. In this paper, we discuss the design and implementation of LegoDB [5], a colonial XML data management system. In particular, we present the complete