Optimization of Distributed Queries

Abstract

Chapter 7 shows how a calculus query expressed on global relations can be mapped into a query on relation fragments by decomposition and data localization. This mapping uses the global and fragment schemas. During this process, the application of transformation rules permits the simplification of the query by eliminating common subexpressions and useless expressions. This type of optimization is independent of fragment characteristics such as cardinalities. The query resulting from decomposition and localization can be executed in that form simply by adding communication primitives in a systematic way. However, the permutation of the ordering of operations within the query can provide many equivalent strategies to execute it. Finding an “optimal” ordering of operations for a given query is the main role of the query optimization layer, or optimizer for short.