John Miles Smith

Database abstractions: aggregation and generalization

Two kinds of abstraction that are fundamentally important in database design and usage are defined. Aggregation is an abstraction which turns a relationship between objects into an aggregate object. Generalization is an abstraction which turns a class of objects into a generic object. It is suggested that all objects (individual, aggregate, generic) should be given uniform treatment in models of the real world. A new data type, called generic, is developed as a primitive for defining such models. Models defined with this primitive are structured as a set of aggregation hierarchies intersecting with a set of generalization hierarchies. Abstract objects occur at the points of intersection. This high level structure provides a discipline for the organization of relational databases. In particular this discipline allows: (i) an important class of views to be integrated and maintained; (ii) stability of data and programs under certain evolutionary changes; (iii) easier understanding of complex models and more natural query formulation; (iv) a more systematic approach to database design; (v) more optimization to be performed at lower implementation levels. The generic type is formalized by a set of invariant properties. These properties should be satisfied by all relations in a database if abstractions are to be preserved. A triggering mechanism for automatically maintaining these invariants during update operations is proposed. A simple mapping of aggregation/generalization hierarchies onto owner-coupled set structures is given.

Database abstractions: aggregation

Aggregation is introduced as an abstraction which is important in conceptualizing the real world. Aggregation transforms a relationship between objects into a higher-level object. A new data type, called aggregate, is developed which, under certain criteria of “well-definedness,” specifies aggregation abstractions. Relational databases defined as collections of aggregates are structured as a hierarchy of n-ary relations. To maintain well-definedness, update operations on such databases must preserve two invariants. Well-defined relations are distinct from relations in third normal form. It is shown that these notions are complementary and both are important in database design. A top-down methodology for database design is described which separates decisions concerning aggregate structure from decisions concerning key identification. It is suggested that aggregate types, and other types which support real-world abstractions without introducing implementation detail, should be incorporated into programming languages.

Optimizing the performance of a relational algebra database interface

An approach for implementing a “smart” interface to support a relational view of data is proposed. The basic idea is to employ automatic programming techniques so that the interface analyzes and efficiently refines the high level query specification supplied by the user. A relational algebra interface, called SQUIRAL, which was designed using this approach, is described in detail. SQUIRAL seeks to minimize query response time and space utilization by: (1) performing global query optimization, (2) exploiting disjoint and pipelined concurrency, (3) coordinating sort orders in temporary relations, (4) employing directory analysis, and (5) maintaining locality in page references. Algorithms for implementing the operators of E. F. Codds relational algebra are presented, and a methodology for composing them to optimize the performance of a particular user query is described.

The design of a rotating associative memory for relational database applications

The design and motivation for a rotating associative relational store (RARES) is described. RARES is designed to enhance the performance of an optimizing relational query interface by supporting important high level optimization techniques. In particular, it can perform tuple selection operations at the storage device and also can provide a mechanism for efficient sorting. Like other designs for rotating associative stores, RARES contains search logic which is attached to the heads of a rotating head-per-track storage device. RARES is distinct from other designs in that it utilizes a novel “orthogonal” storage layout. This layout allows a high output rate of selected tuples even when a sort order in the stored relation must be preserved. As in certain other designs, RARES can usually output a tuple as soon as it is found to satisfy the selection criteria. However, relative to these designs, the orthogonal layout allows an order of magnitude reduction in the capacity of storage local to the search logic.

A Data Base Approach to Software Specification

A data base is a simulation of some real-world phenomena that is represented on a computing machine. A long-standing goal of data base technology has been to specify the semantics of the simulation while suppressing all details of its computer representation. To meet this goal, a variety of specification languages called data base models have been developed. These models are oriented towards applications where large quantities of intricately interrelated data are manipulated by multiple users.

Data base abstraction

Real-world systems modelled by data bases are often quite complex. It is essential that the data base be structured in a way which supports a users abstractions about the real-world system. The “relation” data type is introduced as a structuring primitive for relational data bases The relation data type has similar abstraction properties to the “record” data type of PASCAL. A data base defined via relation data types is structured as a hierarchy of n-ary relations. It is shown that the consistency of this hierarchic structure can be preserved during update operations by two simple rules. A top-down methodology for data base design with relation data types is developed. This methodology minimizes the number of details with which a designer must contend at one time. In particular, the methodology allows the separation of decisions concerning abstract structure from decisions concerning key identification. The separation simplifies both types of decision making for the data base designer.

Database Abstractions: Aggregation and Generalization

Two kinds of abstraction that are fundamentally important in database design and usage are defined. Aggregation is an abstraction which turns a relationship between objects into an aggregate object. Generalization is an abstraction which turns a class of objects into a generic object,. It is suggested that all objects (individual, aggregate, generic) should be given uniform treatment in models of the real world. A new data type, called generic, is developed as a primitive for defining such models. Models defined with this primitive are structured as a set of aggregation hierarchies intersecting with a set of generalization hierarchies. Abstract objects occur at the points of intersection. This high level structure provides a discipline for the organization of relational databases. In particular this discipline allows: (i) an important class of views to be integrated and maintained; (ii) stability of data and programs under certain evolutionary changes; (iii) easier understanding of complex models and more natural query formulation; (iv) a more systematic approach to database design; (v) more optimization to be performed at lower implementation levels. The generic type is formalized by a set of invariant properties. These properties should be satisfied by all relations in a database if abstractions are to be preserved. A triggering mechanism for automatically maintaining these invariants during update operations is proposed. A simple mapping of aggregation/ generalization hierarchies onto owner-coupled set structures is given.

Abstraction in databases

This paper surveys current research and practice concerning abstraction in database systems. Classical and semantic database models are reviewed and emphasized, as fundamental database abstraction mechanisms.