OLAP Query Processing via Subcubes

Abstract

OLAP servers are increasingly being used by many organizations to provide multi-dimensional views which allow a better understanding of data for analysis purposes and provide better performance for analytical queries. Such OLAP queries are complex and time-consuming and hence materializing data cube is a commonly used technique to reduce response time. To our knowledge, most previous OLAP cube implementation techniques apply a static view selection algorithm on the search lattice. These static methods first treat each node in the lattice as an undividable unit and then pick some of them for materialization. Pre-computing some nodes without being aware of which nodes are actually accessed at run time would seriously impact both response time and available space. We propose to further partition nodes in the lattice into subcubes into each of which multiple OLAP queries via a dynamic materialization algorithm can be mapped. Experiments show that the locality effects do exist in OLAP queries, and our dynamic method keeps a reasonable performance even though the available space is very limited and is practical for OLAP query processing.