PhDOOS Logo

Supported by AITO.

These pages are currently maintained by Alex Potanin

 

A Methodology for the Design of Distributed Object DBMS

Fernanda Baiao
( Brazilian )

Federal University of Rio de Janeiro
14 High Point Oaks Ln, apt 302 Madison, WI 53719
tel: 1 608 828 9882
fax: 1 608 265 9777
fernanda@cs.wisc.edu

Keywords:

analysis/design databases distribution Horizontal Fragmentation Vertical Fragmentation

Abstract:

The performance of applications on Object Oriented Database Management Systems (OODBMS) can be significantly improved by the Distribution Design, which reduces irrelevant data accessed by applications and data exchenage among sites in the network. However, in order to improve performance as much as possible, it is very important to design information distribution properly. Since the object data model is much richer than the relational data model, the difficulty of the distribution design task in object databases is significantly increased, because there are several new issues to be taken into consideration.

This research focus on understanding the difficulties involved in the fragmentation of object-oriented databases, evaluating different strategies to handle the problem through experimental studies on top of relevant applications such as the 007 Benchmark, proposing a set of heuristics to address the problem based on those experimental results, and finally developing a unique algorithm to assist distribution designers in the fragmentation phase of the design of distributed object-oriented databases. The proposed algorithm addresses specific characteristics of OODBMS, such as the management of class extensions and object relationships, and it indicates the most adequate fragmentation technique for each class in a database schema, given a set of quantitative and qualitative information about the database structure and about the applications running on it. The major contributions of the algorithm are: (i) it observes performance issues, by allowing a class which has a very small extension not to be fragmented, (ii) it combines vertical, horizontal - primary and derived - and hybrid fragmentation of a class, thus benefiting from the advantages of each technique, and (iii) it permits specific OO characteristics to drive the fragmentation decision process, based on the relationships between classes and on the execution frequency of applications.

We also point to some future works towards (i) an alternative evaluation of the algorithm using a machine learning technique to refine the algorithm, and (ii) a tool to be used by distribution designers in which they will be able to find the most suitable fragmentation schema for a scenario through the proposed algorithm, and experiment alternative fragmentation schemas while evaluating their impact on the application execution time according to a function cost.

The PhD work started: March 1997

The submitted work will probably not be presented in the
upcoming ECOOP PhD Workshop.

HTML3
JAVA

 

Last modified on Mon Aug 15 14:59:24 2005