Querying XML in Timber

In this paper, we describe the TIMBER XML database system implemented at University of Michigan. TIMBER was one of the first native XML database systems, designed from the ground up to store and query semi-structured data. A distinctive principle of TIMBER is its algebraic underpinning. Central contributions of the TIMBER project include: (1) tree algebras that capture the structural nature of XML queries; (2) the stack-based family of algorithms to evaluate structural joins; (3) new rule-based query optimization techniques that take care of the heterogeneous nature of the intermediate results and take the schema information into consideration; (4) cost-based query optimization techniques and summary structures for result cardinality estimation; and (5) a family of structural indices for more efficient query evaluation. In this paper, we describe not only the architecture of TIMBER, its storage model, and engineering choices we made, but also present in hindsight, our retrospective on what went well and not so well with our design and engineering choices. Figure 1: TIMBER Architecture: XML documents are parsed and nodes stored individually in the back-end store. Parsed queries, from multiple supported interfaces, go through a query optimizer to the query evaluator in a relatively standard overall database system architecture. The TIMBER system [10, 16] was developed at the University of Michigan, Ann Arbor, beginning 1999. It was an early native XML data management system. In this retrospective, we take stock of our work over the past nine years. Figure 1 provides an overview of the major system components. Secs. 1 through 4 describe the underlying algebra, query evaluation methods, query optimization, and indices, respectively. Sec. 5 mentions aspects of TIMBER not included in this article. Sec. 6 concludes with a retrospective view. 1 Algebra Relational algebra has been a crucial foundation for relational database systems, and has played a large role in enabling their success. A corresponding XML algebra for XML query processing has been more elusive, due to the comparative complexity of XML, and its history. In the relational model, a tuple is the basic unit of operation and a relation is a set of tuples. In XML, a database is often described as a forest of rooted node-labeled trees. Hence, for the basic unit and central construct of our algebra, we chose an XML query pattern (or Copyright 2008 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE. Bulletin of the IEEE Computer Society Technical Committee on Data Engineering