Overcoming the scalability limitations of parallel star schema data warehouses

Abstract

Most Data Warehouses (DW) are stored in Relational Database Management Systems (RDBMS) using a star-schema model. While this model yields a trade-off between performance and storage requirements, huge data warehouses experiment performance problems. Although parallel shared-nothing architectures improve on this matter by a divide-and-conquer approach, issues related to parallelizing join operations cause limitations on that amount of improvement, since they have implications concerning placement, the need to replicate data and/or on-the-fly repartitioning. In this paper, we show how these limitations can be overcome by replacing the star schema by a universal relation approach for more efficient and scalable parallelization. We evaluate the proposed approach using TPC-H benchmark, to both demonstrate that it provides highly predictable response times and almost optimal speedup. © 2012 Springer-Verlag.