Parallel SQL on Tandem´s NonStop SQL

A comparison is made of a number of different strategies for the concurrent execution of multiple SQL (structured query language) statements on a completely nonsharing multiprocessor system. The author gives some preliminary results in the form of heuristics derived from measurements on a Tandem system. The measurements indicate that in such an environment in most cases the single evaluation of common subexpressions offers no benefits over concurrent execution. Two heuristics have been identified for the evaluation of two frequent structures. It is shown that vertical partitioning performs better than horizontal partitioning of query trees, and that selections should be pushed down to common subtrees.<>