Implications of certain assumptions in database performance evauation

Abstract

The assumptions of uniformity and independence of attribute values in a file, uniformity of queries, constant number of records per block, and random placement of qualifying records among the blocks of a file are frequently used in database performance evaluation studies. In this paper we show that these assumptions often result in predicting only an upper bound of the expected system cost. We then discuss the implications of nonrandom placement, nonuniformity, and dependencies of attribute values on database design and database performance evaluation.