A generalization of the design process for implementing the differential file database architecture
Timothy R. Hill, Jeffrey A. Hoffer · 1988
The differential file architecture (DFA) has been proposed as a basic physical design which enhances the effectiveness of large, active databases. However, a number of performance issues complicate the implementation of the DFA in an actual database environment. There exists a parameter selection procedure for use in implementing the DFA, but its applicability to actual database environments is limited. The procedure is designed to minimize total operating costs, but this requires a prediction of the rate of filtering errors over time for each given set of parameter values being considered. Currently this prediction is provided by a filtering error rate model which assumes an ideal environment in two respects: (1) complete uniformity in the distribution of updates and accesses across data file records, and (2) ideal hashing transformations which uniformly map the record keys to an array of filter bit switches. In several published papers, these assumptions have been challenged as unrealistic for many actual database environments. In support of the Database Administrator (DBA) and the DFA implementation process, the research reported here addresses the following question: What is a generalization of the DFA design process, including a design procedure and support for its use, which is not bound by assumptions of an ideal environment, but rather incorporates the characteristics of specific database environments to yield more accurate design parameter specifications and thus allow more effective implementation? Specifically, the reported research presents: (1) the analytical development of a general model of filtering error rate which accounts for violations of the ideal assumptions, (2) an analysis of computational issues, producing a feasible implementation of the general model as a component of a generalized version of the design procedure, and (3) an experimental evaluation of the generalized procedure, revealing that violations of the ideal assumptions can affect design parameter solutions by as much as 50 percent in some cases. A discussion of these results identifies implications and guidelines for application of the generalized procedure to realistic problems. The resulting generalized design process supports the DBA in implementing the DFA, encouraging the use of this worthwhile innovation in actual database environments.