On the Performance of an Array-Based ADT for OLAP Workloads

There is currently a debate among OLAP vendors on the best way to provide OLAP functionality: Relational OLAP (ROLAP) vendors advocate using sophisticated front ends to provide a multidimensional view of a standard relational database, whereas Multidimensional OLAP (MOLAP) vendors provide custom servers that generally store their data as arrays (instead of tables.) An important question in this debate is the relative performance of arrays vs. tables for common OLAP operations. To shed some light on this question, we have implemented a MOLAP Abstract Data Type (ADT), which uses a multidimensional array as its principle storage mechanism, within the Paradise object-relational DBMS. Using this implementation, we have studied MOLAP and ROLAP performance on the same DBMS platform (by using either the MOLAP ADT or standard relational tables.) In particular, we have compared a new MOLAP consolidation algorithm with three ROLAP consolidation algorithms, and found that for a wide range of data densities, our MOLAP implementation signiicantly outperforms ROLAP both in terms of disk storage and query execution time. This suggests that the ADT mechanism of object-relational database systems is useful for constructing hybrid solutions that provide the beneets of both ROLAP and MOLAP systems.