Estrutura de indexação em memória para dados métricos

Abstract

The stored data recovery in Data bases, in general is made using indexation structures, that allow to much more quickly make the data recovery of that if the search was made sequentially. However, the indexation structures that can be used depend on the domains properties of the of indexing data and the type of queries that must be answered. Traditionally, the data bases management systems support well domains dataset that possess the property of relation of total order, such as numbers and texts with the lexicographical relation order allowing to queries for equality and queries involving order relations such as >, and < Moreover, the commonly indexing structures used in data bases management systems are constructed to be stored in record, partitioning the data set in registers of so great fixture. The example most common of this arrangement is of the indexing trees, when the registers then are called nodes. Sophisticated applications more frequently present datasets in other domains, with other types of quey. When the applications deal with data in metric domain, beyond the proper data elements, a similarity function (or distance) between pairs of elements is defined, and this function is the only way of comparison enters two elements of data of the set. There are diverse metric domain data indexing structures created. However these structures are static (hindering that new elements are added later that the structure was created), or are for record storage. In this work a new dynamic metric structure for metric data total stored in principal memory e was developed. One another interesting property of this structure is that the execution of a quey for existence (point query) covers an only way of search. This characteristic is very interesting, therefore ali the other existing dynamic trees require that the navigation is made not only in depth, but also in width. The structure proposal allows the navigation only in depth for the quey for existence.