ROX: Relational Over XML Alan Halverson, Vanja Josifovski, Guy M. Lohman, Hamid Pirahesh, and Mathias Moerschel   View Paper (PDF)   Return to RESEARCH SESSION 7:XML AND RELATIONS Abstract An increasing percentage of the data needed by business applications is being generated in XML format. Storing the XML in its native format will facilitate new applications that exchange business objects in XML format and query portions of XML documents using XQuery. This paper explores the feasibility of accessing natively-stored XML data through traditional SQL interfaces, called Relational Over XML (ROX), in order to avoid the costly conversion of legacy applications to XQuery. It describes the forces that are driving the industry to evolve toward the ROX scenario as well as some of the issues raised by ROX. The impact of denormalization of data in XML documents is discussed both from a semantic and performance perspective. We also weigh the implications of ROX for manageability and query optimization. We experimentally compared the performance of a prototype of the ROX scenario to today's SQL engines, and found that good performance can be achieved through a combination of utilizing XML's hierarchical storage to store relations "pre-joined" as well as creating indices over the remaining join columns. We have developed an experimental framework using DB2 8.1 for Linux, Unix and Windows, and have gathered initial performance results that validate this approach. ©2005 Association for Computing Machinery