DBC Software Requirement for Supporting Relational Databases.

This is the final report of a series of work aimed at demonstrating the capabilities of a back-end database computer DBC in supporting known data models and systems. In the previous two reports, it was shown that existing hierarchical and network database management systems, in particular, the Information Management System IMS of IBM and DMS1100 of UNIVAC, can be supported on the DBC with a vastly improved performance. In this final report, we study a relational database management system, namely System R, with a view to supporting such a system on the DBC. The early sections of this report are introductory in nature. The representation of relational tuples in the DBC is quite straightforward. The data items of every tuple are converted to attribute-value pairs to form a single DBC record. Two special attribute-value pairs are also included in each DBC record in order to indicate the relation to which the corresponding tuple belongs, and to provide certain clustering information. User transactions in the data sublanguage, called SEQUEL, are converted to a series of DBC commands. The commands are so structured that the DBC can simultaneously access a number of records, the contents of which satisfy the predicates in a SEQUEL query. Given a particular command, the DBC uses its directory to determine the portions of its secondary storage that need to be content-searched.