Cooking DBMS Operations using Granular Primitives

Abstract

The increasing heterogeneity of the underlying hardware forces modern database system engineers to implement multiple variants of a single database operator (e.g., join, selection). With increasing heterogeneity, these variants become too complex to maintain and tune for different devices. To overcome these disadvantages, developers use an alternative, primitive-based operator design. This design paradigm splits the database operators into granular functions or primitives and executes a given operator by combing the necessary primitives. Hence, we require only a limited set of these primitives as we reuse them for multiple database operations. Thus, tuning a single primitive improves efficiency of all the database operations using it. In this survey, we provide an overview of a primitive-based database engine. First, we list different primitives from literature and place them in a hierarchy from the finest granular level to a complete database operator. Second, for each of primitive we list its possible tuning opportunities. Finally, we discuss the significance of primitive-based execution on the query engine. Overall, this survey aims to serve as a general reference for implementing a primitive-based query engine and possible strategies to tune it for specific processors.