Conceptual and relational data models of online transaction processing (OLTP) applications are usually created and maintained following the principle of normalization, which implies avoidance of redundancy. Data retrieval from a disk-based normalized relational database often requires complex and inefficient queries that may cause noticeable performance issues when executed on larger volumes of data. Computer professionals sometimes intentionally trade off the strict normal form to optimize data retrieval queries through error-prone manual tuning and denormalization. We propose a fully automatic optimization approach, based on data redundancy, that relies on a formal cost-benefit model. We prove that finding the optimal level of data redundancy, for given workload statistics, is an NP-Complete optimization problem. A detailed reduction of the problem to binary linear programming is presented in the paper. The proposed optimization approach was evaluated using the TPCE benchmark for OLTP systems. The evaluation has shown that the proposed optimization approach is highly scalable, and that it can be efficiently applied to real-life relational data models.

We outline a methodology for automatic and efficient object-relational mapping (ORM) in the context of model-driven development (MDD) of high-performance information systems specified with executable UML models. Although there are various approaches to performance tuning, we focus here on the persistence layer the relational database. The relational data model is usually designed following the well-known normal forms. However, a fully normalized relational model often does not provide sufficient performance, and improper relational model design can easily lead to a slow and unusable relational database for particular operations. Our ORM approach is intended to exploit smart optimization techniques from the relational paradigm that abandon normalization and its positive effects, and trade them off for better performance. Our ORM approach hence combines the classical denormalization transformations, based on reducing or eliminating expensive database operations by the model restructuring, but applies them to a non-redundant conceptual UML model. In this paper, we also present the first step towards this goal: a catalogue of ORM transformation patterns.