The Object-Slicing Technique: A Flexible Object Representation and Its Evaluation

Recently much work has been done towards extending object-oriented database systems (OODBs) with advanced tools such as view technology, advanced schema evolution support, and role modeling systems. These extensions all require that the underlying database system supports more powerful and exible modeling constructs than are currently supported by existing OODB systems. In this paper, we identify these features as multiple classiication, dynamic reclassiication, and dynamic restructuring. We then describe a methodology known as object-slicing that is capable of extending data models to support these required features. We have successfully implemented an object-slicing software layer using the GemStone system, which while still providing full access to all GemStone DBMS functions, now also ooers all required modeling features. In this paper, we describe our experimental results evaluating the relative costs and beneets of adopting the object-slicing technique. This includes an analytical assessment of the storage overhead of the object-slicing representation, and its comparison against the conventional representational models. As clustering is critical to optimizing queries on such models, we present the results of using OO7 benchmark test suites evaluating various clustering strategies for the object-slicing model. We nd that for certain types of queries (e.g., those that beneet from the superior blocking factor at the local attribute level resulting from clustering object-slices together by class), the object-slicing model outperforms the conventional approach in spite of its storage overhead, while queries involving inherited attributes with low selectivity are better serviced using a conventional object-clustering approach.