Schema Evolution in Object Databases: Measuring the Performance of Immediate and Deferred Updates

When the schema of an object database system is modiied, the database needs to be changed in such a way that the schema and the database remain consistent with each other. This paper uses the OO1 benchmark 2], appropriately modiied, to compare the two most used approaches for transforming the database, namely the immediate and the deferred database transformation 4]. 1 Immediate vs. Lazy Database Updates When the schema of an object database system is modiied, the database needs to be changed in such a way that the schema and the database remain consistent with each other. Several object database systems ooer mechanisms to update the database: in few systems like O 2 3], ObjectStore 10], and OTGenn7] the designer has the possibility to program so called user-deened conversion functions which are associated to modiied classes in the schema and deene how objects have to be restructured. If no conversion functions are provided by the designer, the system provides default conversion functions where no programming is required. Instead, default transformation rules are applied to objects of modiied classes. Systems like Itasca 6] and Versant 12] provide only default transformation rules for transforming the database. No matter if the user intervention is permitted by the system, the main issue considered in this paper is when such conversion functions are executed, that is when the database is brought up to a consistent state wrt. the new schema. There are mainly two approaches for implementing database conversion functions: immediate and deferred 4]. In the rst case, all objects of the database are updated immediately as soon as the schema modiication is performed, whereas with the deferred approach objects are updated only when they are actually used. This paper focuses on a performance evaluation of the two mentioned approaches and identiies those parameters that innuenced most the results. 2 Experimental Results In this section we present some experimental results obtained making use of OSLOT (Object Storage with Lazy Object Transformations), an object storage system developed at the University of Frankfurt particularly implemented for schema and database evolution testing and comparison 9]. After a schema modiication, OSLOT updates the database using the algorithms deened in 4]. Section 2.1 describes the benchmark organization and Section 2.2 presents the benchmark results with respect to the immediate and the deferred database transformation. 2.1 The Benchmark Organization OSLOT is developed on top of EOS 1] using Tcl/Tk 11] and …