Multiple-query optimization

Reviewer: Edward Sava-Segal

Given a set of queries that are supposed to share some tasks, a considerable reduction of execution time may be achieved by avoiding redundant page access. Sharing common temporary results among queries is cheaper than serial execution. This is the main reason for performing multiple-query analysis. The author's concern is building an optimal access plan, given a set of queries and a set of common subexpressions found previously. Two architectures for systems with a multiple-query processing facility are considered. One is based on a plan merger interleaving the results of the locally optimal access plans of each query. A second more sophisticated one is based on a global optimizer. In the second case the optimization problem is formulated as a state space search problem, and an admissible algorithm is used for that problem. Compared to previous work, a better estimation function is defined in this paper. It penalizes plans that cannot coexist in the final state, thus enabling the algorithm to converge to the final solution faster. Some restrictions are made in building the algorithm, but a way to avoid them using the same framework is discussed. An overview of previous research in the area is presented, and a full section describes experimental results, which the author claims to be publishing for the first time. Experiments proved the usefulness of the given algorithms in various contexts, producing a decrease of 20–50 percent in both I/O and CPU time, even in the presence of fast access paths for relations. Algorithms are clearly described and exemplified. No previous background, except some acquaintance with heuristic search algorithms, is needed to fully understand this interesting paper. Everyone concerned with building database and rule-based systems should read it.