Using genetic algorithms to work out index configurations for the class-hierarchy indexing in object databases

Ef®cient indexing on a class hierarchy is essential for the achievement of high performance in query evaluation for object databases. In this paper, we present a practical indexing scheme, Partition Index Con®guration Scheme (PINS), which provides good index con®gurations for any real database environment. PINS considers the distribution of key values, as well as query patterns such as query frequency on each class. PINS can easily be applied to any database system, since it uses the B 1-tree structure. We develop a cost model and, through experiments, demonstrate the performance of the proposed policy over various class hierarchies. q 2000 Elsevier Science B.V. All rights reserved.