Database Systems: Issues in the Design of a Distributed Record Management System

Issues in the Design of a Distributed Record Management System By J. P. LINDERMAN (Manuscript received September 15, 1981) Inexpensive processors will lead to powerful new tools for constructing software systems. However, the introduction of intelligent hardware may be limited by software architecture. If intelligence has been stripped away in the process of top-down refinement, new devices will be constrained to behave like their unintelligent predecessors. Software design will have to be flexible in distributing intelligence to accommodate processor-based tools gracefully. I discuss a conventional design for a record-retrieval facility. Some features that are acceptable and even attractive in a single-processor environment are shown to limit use of multiple processors and smart peripherals. I propose a less conventional design that can exploit special-purpose hardware and provide a smooth growth path from single-processor systems. I. INTRODUCTION The ideas in this paper are a result of my participation in the design of an experimental data-management system. The goal of the design is less to produce a database manager than to produce a system of components from which a database manager, or many database managers, can be constructed. Major components will have several interchangeable implementations exhibiting a range of price and performance. The design team hopes to assemble individual database managers by matching the various components to the requirements of the database system.