Extensibility in the starburst: Experimental database system

Abstract

The Starburst experimental database system is a research tool for experimenting with advanced database implementation techniques. As such, it is important to be able to extend the system in a variety of ways to facilitate the evaluation of the design and implementation of various advanced database management functions. In addition, many applications that use (or wish to use) database systems to store their persistent state can benefit from specialized constructs or facilities, either for better performance, or to model more closely the application domain. The conflicting goals of application-specific facilities and integration of information across application boundaries are (perhaps) best addressed by a database system supporting domain-specific extensions in the context of a common data model. A common data model, with its accompanying data access and manipulation interfaces facilitates information sharing between applications. Domain-specific extensions allow the system to be adapted to support specialized data management and modeling requirements. Domain-specific extensions, in Starburst, range from data storage extensions to support application specific requirements (e.g., main memory storage or spatial searching), to data modeling mechanisms allowing the specification of data semantics (behavior and constraints) in terms of the primitive concepts of the application domain. The common data model of Starburst is the relational data model [CODD 70] and the data access and manipulation interface is an extension of SQL [CHAMBERLIN 76, SQL]. Starburst extensibility begins with the ability to support alternative methods for storing and accessing relations. It continues with facilities for extending the repertoire of execution algorithms for evaluating relational expressions. Additionally, Starburst extends the range of relational expressions to include recursive and structured-result queries over stored and derived tables. Finally, the Starburst project will explore mechanisms for enriching the semantics of stored data with a data type definition facility, data monitoring mechanisms (e.g., triggered rules & integrity constraints), and a facility for defining and manipulating complex (composite) objects. This article reviews the mechanisms in Starburst that allow database researchers and developers of domain-specific applications to modify the system to test new approaches or to address application requirements. A common theme in the Starburst approach to extensibility is the tension between general-purpose constructs and the need for specialized constructs that sometimes do not match the general case. In particular, specialized functions often require specialized parameterization. We discuss the general facilities and the mechanisms for their specialization. Data storage, query analysis, and query evaluation all support extensions that refine the general construct and allow extension-specific handling at various points.