This paper covers a lot of territory in its 16 pages, and it covers it exceptionally well. Briefly, the paper presents the details of the design, execution, and analysis of a benchmark test between a conventional database system (ORACLE using SQL running on a VAX 11/750 with VMS 3.0 and a 9766 CDC disk drive) and a database machine configuration (Britten-Lee IDM-500 using QUEL connected to a VAX 11/750 front end with Berkeley UNIX 4.1 and a 9766 CDC disk drive for the IDM).

The database on each system was constructed from data from an actual application database, and the test databases each had 12 relations in the relational scheme. The researchers developed ten primary query sets and several special-purpose query sets to test specific database features. Each query set contained from four to seven queries of varying complexity. Specific sets were devised to cover single relation and multiple relation (join) queries.

The benchmark test and the analyses of the results appear to be exceptionally comprehensive and complete. Whereas many similar papers concentrate on one or two variables, the authors of this paper have covered all relevant variables and have presented, interpreted, and explained the results. There appear to be few, if any, assumptions and no apparent researcher biases in the tests or analysis.

The authors have presented a benchmark and results that are as objective as possible. The results, as well as the authors’ interpretation or explanation of them, exhibit a comprehensive understanding of the pertinent factors that could affect such a study and a rationale, a methodology, and an execution that are difficult to find fault with.

Overall, the benchmark results show the clear advantage of the database machine architecture over the conventional database system architecture. The authors have taken pains to minimize any factors that might impede nonarchitectural comparisons such as line speeds. This paper is highly recommended from two perspectives: first, for those looking to determine the relative advantage of either of the two architectures, and second (and perhaps of equal value), as an example of how to design, execute, and analyze a benchmark study.