Automatic Schema Design for Co-Clustered Tables

Schema design of analytical workloads provides opportunities to index, cluster, partition and/or materialize. With these opportunities also the complexity of finding the right setup rises. In this paper we present an automatic schema design approach for a table co-clustering scheme called Bitwise Dimensional Co-Clustering, aimed at schemas with a moderate amount dimensions, but not limited to typical star and snowflake schemas. The goal is to design one primary schema and keep the knobs to turn to a minimum while providing a robust schema for a wide range of queries. In our approach a clustered schema is derived by trying to apply dimensions throughout the whole schema and co-cluster as many tables as possible according to at least one common dimension. Our approach is based on the assumption that initially foreign key relationships and a set of dimensions are defined based on classic DDL.