Abstract

OLAP systems are designed to quickly answer multi-dimensional queries against large data warehouse systems. Constructing data cubes and their associated indexes is time consuming and computationally expensive, and for this reason, data cubes are only refreshed periodically. Increasingly, organizations are demanding for both historical and predictive analysis based on the most current data. This trend has also placed the requirement on OLAP systems to merge updates at a much faster rate than before.
In this thesis, we proposes a framework for OLAP systems that enables updates to be merged with data cubes in soft real-time. We apply a strategy of local partitioning of the data cube, and maintain a ``hot'' partition for each materialized view to merge update data. We augment this strategy by applying multi-core processing using the OpenMP library to accelerate data cube construction and query resolution.
Experiments using a data cube with 10,000,000 tuples and an update set of 100,000 tuples show that our framework achieves a 99% performance improvement updating the data cube, a 76% performance increase when constructing a new data cube, and a 72% performance increase when resolving a range query against a data cube with 1,000,000 tuples.