Physical integrity in a large segmented database

Abstract

A database system can generally be divided into three major components. One component supports the logical database as seen by the user. Another component maps the information into physical records. The third component, called the storage component, is responsible for mapping these records onto auxiliary storage (generally disks) and controlling their transfer to and from main storage. This paper is primarily concerned with the implementation of a storage component. It considers a simple and classical interface to the storage component: Seen at this level the database is a collection of segments. Each segment is a linear address space. A recovery scheme is first proposed for system failure (hardware or software error which causes the contents of main storage to be lost). It is based on maintaining a dual mapping between pages and their location on disk. One mapping represents the current state of a segment being modified; the other represents a previous backup state. At any time the backup state can be replaced by the current state without any data merging. Procedures for segment modification, save, and restore are analyzed. Another section proposes a facility for protection against damage to the auxiliary storage itself. It is shown how such protection can be obtained by copying on a tape (checkpoint) only those pages that have been modified since the last checkpoint.