Optimal MMI file systems for orthogonal range retrieval

Abstract

Range queries are more general and significant than partial match queries (PMQs) in the applications of multi-attribute information retrieval. However, almost all efforts devoted to the design of multi-attribute file systems so far were concentrated on PMQs. In this paper, we are concerned with the problem of designing multi-attribute file systems for facilitating orthogonal range queries (ORQs). A greedy method, called the minimum marginal increase (MMI) method, is presented first. Then we derive the formulas of performance of an important static file and an important dynamic file, namely the multiple key hashing (MKH) file and the extendible hashing (EH) file, for answering ORQs. Based upon these performance formulas, we show that the problem of designing optimal file systems for ORQs is indeed a very difficult problem. Further, we show that the presented MMI method can indeed be used to design optimal MKH files and optimal EH files, in some cases, for ORQs. Finally, some experiments show that the MMI method can still be applied to produce “good” MKH files and “good” EH files for the general case.