Selecting Subexpressions to Materialize for Dynamic Large-Scale Workloads

The nature of analytical queries executed either inside or outside of a DBMS increases the redundant computations due to the presence of common query sub-expressions. More recently, a few largely industry-led studies have focussed on the problem of identifying beneficial sub-expressions for large-scale workloads running outside of a DBMS for materialization purposes. However, these works have unfortunately ignored the large-scale workloads running inside of a DBMS. To align them in terms of the materialization of beneficial sub-expressions for dynamic large-scale workloads, we propose a pro-active approach that uses hypergraphs. These structures exploit cost models towards capturing the common query sub-expressions and materializing the most beneficial ones. Our approach is accompanied by a strategy, which orients the first $\delta$ queries to the offline phase that selects their appropriate views. To augment the benefit and sharing of the selected views, the initial $\delta$ queries may be scheduled. The online phase manages the pool of views obtained by the first phase by adding/dropping views to optimize new incoming queries. We conducted extensive experiments to evaluate the efficiency of our proposal as well as its cost-effective integration in a commercial DBMS.