Distributed query evaluation on semistructured data

Abstract

Semistructured data is modeled as a rooted, labeled graph. The simplest kinds of queries on such data are those which traverse paths described by regular path expressions. More complex queries combine several regular path expressions, with complex data restructuring, and with sub-queries. This article addresses the problem of efficient query evaluation on distributed, semistructured databases. In our setting, the nodes of the database are distributed over a fixed number of sites, and the edges are classified into local (with both ends in the same site) and cross edges (with ends in two distinct sites). Efficient evaluation in this context means that the number of communication steps is fixed (independent on the data or the query), and that the total amount of data sent depends only on the number of cross links and of the size of the query's result. We give such algorithms in three different settings. First, for the simple case of queries consisting of a single regular expression; second, for all queries in a calculus for graphs based on structural recursion which in addition to regular path expressions can perform nontrivial restructuring of the graph; and third, for a class of queries we call select-where queries that combine pattern matching and regular path expressions with data restructuring and subqueries. This article also includes a discussion on how these methods can be used to derive efficient view maintenance algorithms.