Mathematical Techniques for Efficient Record Segmentation in Large Shared Databases

Abstract

It is possible to significantly reduce the average cost of information retrieval from a large shared database by partitioning data items stored within each record into a primary and a secondary record segment. An analytic model, based upon knowledge of data item lengths, transportation costs, and retrieval patterns, is developed to assist an analyst with this assignment problem. The model is generally applicable to environments in which a database resides in secondary storage, and is useful for both uniprogramming and multiprogramming systems. A computationally tractable record design algorithm has been implemented as a Fortran program and applied to numerous problems. Realistic examples are presented which demonstrate a potential for reducing total system cost by more than 65 percent.