Query Processing Techniques for Multiversion Access Methods

Abstract

Multiversion access methods have been emerged in the literature primarily to support queries on a transaction-time database where records are never physically deleted. For a popular class of efficient methods (including the multiversion Btree), data records and index entries are occasionally duplicated to separate data according to time. In this paper, we present techniques for improving query processing in multiversion access methods. In particular, we address the problem of avoiding duplicates in the response sets. We first discuss traditional approaches that eliminate duplicates using hashing and sorting. Next, we propose two new algorithms for avoiding duplicates without using additional data structures. The one performs queries in a depth-first order starting from a root, whereas the other exploits links between data pages. These methods are discussed in full details and their main properties are identitied. Preliminary performance results confirm the advantages of these methods in comparison to traditional ones according to CPU-time, disk accesses and storage.