A graph-based decomposition approach for recursive query processing

Abstract

In practice most recursive logic queries to a deductive database are linear. According to /JaAgNe 87/ every linearly recursive query can be expressed by one so-called transitive closure query and several non-recursive queries.

A transitive closure query TC refers to two binary relations R and Q. It corresponds to a so-called database digraph DG=(V_R,R) and a set V_goal of special nodes in DG. Given an arbitrary c ∈ V_R. The major step in the bottom-up evaluation of TC is to efficiently compute the set of all (c,Y) ∈ R⋆, Y ∈ V_goal.

The first technique clusters the database digraph, which yields a connection hypergraph CH, a clustergraph CG and a skeleton graph SG. CG is the homomorphic image of DG w.r.t. the clustering, SG is a small special subgraph of the transitive closure DG⋆=(V_R,R⋆) of DG. Based on CG and CH the query TC is decomposed into auxiliary queries that refer only to the relevant parts of R. This reduces the amount of intermediate tuples generated during query evaluation. Augmenting the database by the edges of the skeleton graph speeds up the evaluation of the auxiliary queries for arbitrary constants ‘c’.

The second technique refines the decomposition approach by a further clustering technique using some special knowledge about the structure of R. Every cluster contains one prototypical element, such that it suffices to solve auxiliary queries for the prototypes, since the results for other elements can then be derived easily from the prototypical results.

The combination of both clustering techniques is implemented in a Prolog-based advanced deductive database system, that solves traversal recursion problems on very large relations R, e.g. computes connecting paths in DG.