Dynamic Pre-fetching of Views Based on User-Access Patterns in an OLAP System

Materialized view selection plays an important role in improving the efficiency of an OLAP system. To meet the changing user needs, many dynamic approaches have been proposed for solving the view selection problem. Most of these approaches use some form of caching to store frequently accessed views and a replacement policy to replace the infrequent ones. While some of these approaches use on-demand fetching, where the view is computed only when it is asked, a few others have used a pre-fetching strategy, where certain additional information is used to pre-fetch views that are likely to be accessed in the near future. In this paper, we propose a global pre-fetching scheme that uses user access pattern information to pre-fetch certain candidate views that could be used for efficient query processing within the specified user context. For specific kinds of query patterns, called drill-down analysis, which is typical of an OLAP system, our approach significantly improves the query performance by pre-fetching drill-down candidates that otherwise would have to be computed from the base fact table. We compare our approach against dynamat; a wellknown on-demand fetching based dynamic view management system that is already known to outperform optimal static view selection. The comparison is based on the detailed cost savings ratio, used for quantifying the benefits of view selection against incoming queries. The experimental results show that our approach outperforms dynamat and thus, also the optimal static view selection.