Authors: Chun-Hung Cheng, Jaideep Motwani
Addresses: Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong, Shatin, N.T., SAR, Hong Kong. ' Department of Management, Seidman College of Business, Grand Valley State University, 301 West Fulton, Suite 409C, Grand Rapids, MI 49504, USA
Abstract: A major objective in database physical design is to minimise disk access in processing database transactions. Vertical partitioning is a major design approach to ensure fewer disk accesses. In this paper, we consider the use of Cluster Identification (CI)-based approaches to produce vertical partitions in database design. To produce partitions, the existing CI-based approaches either remove attributes or duplicate attributes in the branching scheme. It turns out that the former sometimes gives undesirable partitions and that the latter produces overlapping partitions. To overcome these deficiencies, we propose to remove accesses in the branching scheme. Illustrative examples are used to show the effectiveness of this branching scheme. A computational study is also conducted. In addition, the new approach is extended to deal with a vertical partition case considering frequencies of accesses.
Keywords: clustering; cluster identification; branch-and-bound; data partitions; vertical partitions; database design; disk access.
International Journal of Business Information Systems, 2009 Vol.4 No.6, pp.622 - 638
Published online: 23 Jun 2009 *Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article