Logical Schema for Data Warehouse on Column-Oriented NoSQL Databases

The column-oriented NoSQL systems propose a flexible and highly denormalized data schema that facilitates data warehouse scalability. However, the implementation process of data warehouses with NoSQL databases is a challenging task as it involves a distributed data management policy on multi-nodes clusters. Indeed, in column-oriented NoSQL systems, the query performances can be improved by a careful data grouping. In this paper, we present a method that uses clustering techniques, in particular k-means, to model the better form of column families, from existing fact and dimensional tables. To validate our method, we adopt TPC-DS data benchmark. We have conducted several experiments to examine the benefits of clustering techniques for the creation of column families in a column-oriented NoSQL HBase database on Hadoop platform. Our experiments suggest that defining a good data grouping on HBase database during the implementation of a data warehouse increases significantly the performance of the decisional queries.