Experimental variations of a theoretically good retrieval data structure

Abstract

A retrieval data structure implements a mapping from a set S of n keys to range R={0,1} r , e.g. given by a list of key-value pairs (x,v) S×R, but an element outside S may be mapped to any value. Asymptotically, minimal perfect hashing allows to build such a data structure that needs nlog2 e+nr+o(n) bits of memory and has constant evaluation time. Recently, data structures based on other approaches have been proposed that have linear construction time, constant evaluation time and space consumption O(nr) bits or even (1+ε)nr bits for arbitrary ε>0. This paper explores the practicability of one such theoretically very good proposal, bridging a gap between theory and real data structures. © 2009 Springer Berlin Heidelberg.