Data management and query processing for semistructured data
Jennifer Widom, Jason McHugh · 2000
Traditional database management systems require all data to adhere to an explicitly specified, rigid schema. However, a large amount of the information available today is semistructured—the data may be irregular or incomplete, and its structure may evolve rapidly and unpredictably. It is difficult and inefficient to manage semistructured data using traditional relational, object-oriented, or object-relational database systems, which are designed and tuned for well-structured data. This thesis describes Lore, a new database management system we developed for storing and querying semistructured data. The overall architecture of the Lore system contains many of the traditional database system components, but the fundamentally different nature of schema-less, semistructured data has required new techniques inside each component. This thesis covers our work in the overall system architecture, its query language, access methods, cost-based query optimizer, and view manager. We also describe a mechanism we developed by which Lore can dynamically and invisibly fetch and cache data from external sources during query processing.