Identifier Search Mechanisms: A Survey and Generalized Model

Abstract

Generalized ModelAn important functional component of ever.y file organization is the set of search mechanisms which are used to locate individually identified records during the process of either update or retrieval. There have been a large number of computer-related search techniques developed in the past 20 years. The objective of this article is to synthesize many of these techniques into one parametrically describable search mechanism. This is accomplished by separating search techniques into two categries: addressing techniques, and tree searching techniques. Significant literature is surveyed for each category. The new concept of a TRIE-TREE search mechanism is then introduced, and used to combine these categories via a generalized model. The model has proven to be a pedagogic convenience in explaining the relation between alternative search techniques. It has also been used irk a practical application as a schema for alternative search strategies which are automatically evaluated during the process of a file organization design.

Burst tries

“…A list implementation offers space savings if a trie node has only a few children, when an array of fixed size would consist largely of null pointers. However, the space savings come at the expense of a linked list traversal in each node instead of a simple array lookup [43]. Sussenguth [46] proposed a modified form of the linked list implementation of a trie, called a double-chained tree, in which all children of a node are placed in a list as in the linked-list implementation explained above.…”

Order preserving extendible hashing and bucket tries

“…This together with the unifying approach of [13] motivates our differing method of introducing EXHASH. It also appears that a one level directory is not always optimal for order preserving address transformations (as in the case analyzed in [1]) because of the large effect on directory size caused by non-uniformity.…”

“…For each i, 0 =< i < di(T), there is a corresponding directory interval I ~ dm, aAccess to the leaf buckets of a trie can be implemented either as an (m-ary) tree [13,15,22,24] or utilizing a one level directory with di (T) elements I-1, 13] leading to direct calculated access to the leaves. Knuth [24, exercise 6.3.20] describes a structure very similar to the BT above.…”

“…We regard the new dynamic (extendible) hashing schemes [t, 15, 16, 17, 18, 19] from the practical viewpoint as very relevant implementations of general "parametrizable" ideas [13,14,20,21,22] known quite long. The linear hashing of Litwin [18] and Larson [16] seems most novel in its hash function extension mechanism.…”

“…We call such a data structure a bucket trie (BT). The trie and trie-tree as a generalized search mechanism have been lucidly described in [13]. Coffman and Eve [22] analyze tries assuming uniform randomization and Knuth [24] presents (as exercises) results on the number of leaves in a BT.…”