Finite Cursor Machines in Database Query Processing

Abstract

A database system is often concerned with the processing of lists of tuples in a single scan, using constant amount of memory. In classical relational query processing [2], many of the relational algebra operators have simple single-scan implementations on sorted lists. In more recent data stream systems [1], single-scan processing is a must. Data warehousing software tools, such as those by Aruna, support database querying using index structures for text searching.

To improve our understanding of the possibilities and limitations of single-scan, constant-memory processing on lists of tuples, we define and study the abstract model of finite cursor machines. Finite cursor machines are, of course, instantiations of sequential ASMs.

In conjunction with sorting, finite cursor machines can evaluate a wide class of relational algebra expressions; in particular, they can compute all database queries expressible using semijoins rather than full joins. Challenging problems include delineating the precise computing power of finite cursor machines with sorting, and minimizing the number of sorting operations that are needed. We discuss these problems and present some preliminary results.

This is joint work with Dirk Leinders and Jerzy Tyszkiewicz.