Relational Query Optimization with Enhanced ADTs

Abstract

Object-relational queries access large complex data types and expensive methods of those data types. In earlier work, we modeled complex types as "Enhanced ADTs" (E-ADTs) and demonstrated the resulting performance improvements when implemented in the PREDATOR system. This paper explores the opportunities for further improvements through interactions between relational query optimization and E-ADT optimization. We identify four broad categories of optimization opportunities and study specific examples in each of these categories. These examples span query rewrite, indexing, aggregation and join optimization. Our conclusion is that non-trivial interactions exist between E-ADTs and relational queries, and that special optimization techniques are necessary to achieve good performance. These techniques have been prototyped in PREDATOR, and we present experimental results that demonstrate their effect.