Type-Level access pattern view: a technique for enhancing prefetching performance

• Authors: Wook-Shin Han;Woong-Kee Loh;Kyu-Young Whang
• Venue: DASFAA'06 Proceedings of the 11th international conference on Database Systems for Advanced Applications
• Year: 2006

Navigational applications on Object-Relational DBMSs (ORDBMSs) access objects in the database related to one another via reference and collection attributes. When accessing an object, the applications first look up the object cache in the client and, if the object does not exist, fetch the object from the server. Prefetching is to identify the objects that are highly probable to be accessed in the future by the applications and to save these objects in the object cache in advance. Since prefetching reduces the number of high cost fetches, it is crucial for the performance of the applications. The prefetch method proposed by Han et al.[7] reduces the number of fetches by orders of magnitude compared with the previous methods. However, overall performance enhancement is not as significant as reduction of fetches. Since the performance of prefetching is determined by the number of disk accesses in the server as well as the number of fetches. In this paper, we propose a technique for minimizing the number of disk accesses to enhance the performance of the prefetch method proposed by Han et al. We propose a method for creating materialized views based on the type-level path access logs proposed by Han et al.[6]. We call the materialized view as the type-level access pattern view. We then present an algorithm for minimizing the number of disk accesses when prefetching the objects from the database in the server by using the type-level access pattern view. We perform a series of experiments using a variety of databases to show that the technique proposed in this paper significantly enhances the overall performance of the navigational applications. We show that the proposed technique reduces the number of disk accesses by up to 33.0 times and enhances the performance by up to 21.4 times compared with the original prefetch method by Han et al.