Update and retrieval in a relational database through a universal schema interface

Abstract

A database system that is based on the universal relation (UR) model aims at freeing its users from specifying access paths on both the physical and on the logical levels. All information about the logical structure of the database (i.e., its conceptual scheme) is hidden from users; they need only to know the attribute names, which now carry all the semantics of the database. Previous work on UR interfaces has concentrated on the design and implementation of query languages that serve to facilitate retrieval of data from a relational database. On the other hand, updates are always handled as before, which means that users still have to know the logical structure of the database in case they want to insert, delete, or modify tuples. In this paper the concepts underlying a UR interface, which is really “universal,” are presented; it is based on the UR model, and it permits not only queries but also updates: Combinations of attributes that may participate in an update-operation ("objects") have to be specified during the design phase of the database, and are then embodied into the database scheme by an extended synthesis algorithm. They form the basis for any insertion or deletion operation. A precise definition of “insertable” tuples, and of the insert- and delete-operation in this new context, is given. It is then shown that these operations modify a database state in such a way that a representative instance always exists. This is accomplished by providing a more detailed version of Sagiv's uniqueness condition and by exploring the structure of nonunique objects. Since the underlying database always has a representative instance, this instance can be used to define the window function for retrieval. It is shown that it is still possible to compute windows by a union of minimal extension joins.