Abstract. Horizontal fragmentation plays an important role in the design phase of Distributed Databases. Complex class relationships: associations, aggregations and complex methods, require fragmentation algorithms to take into account the new problem dimensions induced by these features of the object oriented models. We propose in this paper a new method for horizontal partitioning of classes with complex attributes and methods, using AI clustering techniques. We provide quality and performance evaluations using a partition evaluator function and we prove that fragmentation methods handling complex interclass links produce better results than those ignoring these aspects. 1.

Abstract:- Class fragmentation is an essential phase in the design of Distributed Object Oriented Databases (DOODB). Due to their semantic similarity with the purpose of database fragmentation (obtaining sets of similar objects with respect to the user applications running in the system), clustering algorithms have recently begun to be investigated in the process of database fragmentation. This work proposes a study on the impact of different similarity measures applied in hierarchical agglomerative clustering algorithms for horizontal fragmentation of classes with complex attributes. This study would eventually help finding formal, automatic, approaches in choosing a particular similarity measure in accordance with: the applied clustering algorithm, the structure of the database inheritance/aggregation hierarchies, the semantics of data, etc. Key-Words:- Distributed database design, horizontal fragmentation, data mining methods, performance evaluation. As opposed to centralized databases where the design phase handles only logical and physical data modeling, the design process in Distributed Object Oriented Databases involves both data partitioning