Towards the Study of Performance Trade-offs Between Materialized and Virtual Integrated Views

Consider the problem of supporting an integrated view over multiple databases. The traditional approach is to use a virtual view, but recent investigations are proposing to use a materialized view, or a hybrid virtual/materialized view. This paper initiates an investigation into the performance trade-oos along this spectrum of choices. In particular, the paper develops analytical models for predicting query-response time, view freshness, and system load in terms of parameters such as query frequency, query complexity, update frequency, and network delay. In many ways the performance of a mediator-based integration environment is similar to that of a client-server DBMS architecture. However, the notion of query freshness does not explicitly arise in a single DBMS, and complex joins may more likely in an integration environment. This paper lays the groundwork for conducting bench-marking experiments for integrated views, and presents the results of some initial experimentation. Such results will be used to calibrate the analytical model for speciic application environments. This will enable prediction of the behavior of the virtual and materialized approaches in speciic contexts.