Runtime specialization of postgreSQL query executor

Abstract

For computationally intensive workloads, achieving high database performance is in direct correspondence to utilizing CPU efficiently. At the same time, interpretation overhead inherent to traditional interpretive SQL engines gets in the way of optimal CPU utilization. One solution to this problem is dynamic query compilation, which consists in generating efficient machine code at run time given a particular input query. Creating a complete query compiler from scratch for an existing database system takes a large amount of development and maintenance effort. Similar results, however, can be obtained more easily using program specialization of a generic query engine with respect to a particular query. This paper presents intermediate results of applying this approach to the query engine at the core of PostgreSQL database system.